U.S. patent number 8,831,546 [Application Number 13/536,203] was granted by the patent office on 2014-09-09 for mrc antenna diversity for fm iboc digital signals.
This patent grant is currently assigned to iBiquity Digital Corporation. The grantee listed for this patent is Jeffrey S. Baird, Brian W. Kroeger, Paul J. Peyla. Invention is credited to Jeffrey S. Baird, Brian W. Kroeger, Paul J. Peyla.
United States Patent |
8,831,546 |
Kroeger , et al. |
September 9, 2014 |
MRC antenna diversity for FM IBOC digital signals
Abstract
A radio receiver includes a first signal path including a first
tuner configured to receive a first signal from a first antenna,
and a first demodulator configured to demodulate symbols from an
output of the first tuner to produce first branch metrics derived
from the demodulated symbols; a second signal path including a
second tuner configured to receive a second signal from a second
antenna, and a second demodulator configured to demodulate symbols
from an output of the second tuner to produce second branch metrics
derived from the demodulated symbols; a combiner for maximum ratio
combining the first branch metrics and the second branch metrics;
and processing circuitry to process the combined first and second
branch metrics to produce an output signal.
Inventors: |
Kroeger; Brian W. (Sykesville,
MD), Peyla; Paul J. (Elkridge, MD), Baird; Jeffrey S.
(Columbia, MD) |
Applicant: |
Name |
City |
State |
Country |
Type |
Kroeger; Brian W.
Peyla; Paul J.
Baird; Jeffrey S. |
Sykesville
Elkridge
Columbia |
MD
MD
MD |
US
US
US |
|
|
Assignee: |
iBiquity Digital Corporation
(Columbia, MD)
|
Family
ID: |
48224014 |
Appl.
No.: |
13/536,203 |
Filed: |
June 28, 2012 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20130115903 A1 |
May 9, 2013 |
|
Related U.S. Patent Documents
|
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
Issue Date |
|
|
61556428 |
Nov 7, 2011 |
|
|
|
|
Current U.S.
Class: |
455/193.1;
375/267; 375/347; 455/334; 455/137 |
Current CPC
Class: |
H04B
7/0857 (20130101); H04B 7/0845 (20130101); H04B
7/0885 (20130101); H04H 2201/18 (20130101) |
Current International
Class: |
H04B
1/18 (20060101) |
Field of
Search: |
;455/132,137,150.1,193.1,313,323,334 ;375/267,347 |
References Cited
[Referenced By]
U.S. Patent Documents
Other References
US. Appl. No. 13/165,325, filed Jun. 21, 2011. cited by applicant
.
U.S. Appl. No. 13/189,659, filed Jul. 25, 2011. cited by
applicant.
|
Primary Examiner: Le; Nhan
Attorney, Agent or Firm: Lenart, Esq.; Robert P. Pietragallo
Gordon Alfano Bosick & Raspanti, LLP
Parent Case Text
CROSS-REFERENCE TO A RELATED APPLICATION
This application claims the benefit of U.S. Provisional Patent
Application No. 61/556,428, filed Nov. 7, 2011, which is hereby
incorporated by reference.
Claims
What is claimed is:
1. A radio receiver comprising: a first signal path including a
first tuner configured to receive a first signal from a first
antenna, and a first demodulator configured to demodulate symbols
from an output of the first tuner to produce first branch metrics
derived from the demodulated symbols; a second signal path
including a second tuner configured to receive a second signal from
a second antenna, and a second demodulator configured to demodulate
symbols from an output of the second tuner to produce second branch
metrics derived from the demodulated symbols; a combiner for
maximum ratio combining the first branch metrics and the second
branch metrics, wherein the first and second branch metrics are
synchronized by indexing; and processing circuitry to process the
combined first and second branch metrics to produce an output
signal.
2. The radio receiver of claim 1, wherein the indexing occurs prior
to the deinterleaving.
3. The radio receiver of claim 1, wherein indexing is used to
identify the first and second branch metrics by a symbol
number.
4. The radio receiver of claim 1, wherein when one of the signal
paths has no branch metrics available, the branch metrics of that
signal path are zeroed.
5. The radio receiver of claim 1, wherein the first and second
demodulators adjust the magnitudes of the first and second branch
metrics in response to signal-to-noise ratios of the first and
second signals.
6. The radio receiver of claim 1, wherein the combiner sums
corresponding, synchronized branch metrics from the first and
second signal paths.
7. The radio receiver of claim 1, wherein the processing circuitry
includes a deinterleaver and a Viterbi decoder, and wherein
like-indexed branch metrics are added when the corresponding
symbols are available from the first and second signal paths.
8. The radio receiver of claim 1, wherein each of the signal paths
independently acquires and tracks a signal received by one of the
antennas.
9. The radio receiver of claim 8, wherein symbol and frequency
tracking for each signal path flywheels over temporary fades or
outages.
10. The radio receiver of claim 1, wherein the first and second
antennas are configured to receive an FM IBOC signal.
11. The radio receiver of claim 1, wherein the processing circuitry
processes the second branch metrics to produce a data output
signal.
12. A radio receiver comprising: a first signal path including a
first tuner configured to receive a first signal from a first
antenna, and a first demodulator configured to demodulate symbols
from an output of the first tuner to produce first branch metrics
derived from the demodulated symbols; a second signal path
including a second tuner configured to receive a second signal from
a second antenna, and a second demodulator configured to demodulate
symbols from an output of the second tuner to produce second branch
metrics derived from the demodulated symbols; a combiner for
maximum ratio combining the first branch metrics and the second
branch metrics; processing circuitry to process the combined first
and second branch metrics to produce an output signal; a third
signal path including a third tuner configured to receive the
second signal from the second antenna, and a third demodulator
configured to demodulate symbols from an output of the third tuner;
and processing circuitry to process an output of the third
demodulator to produce a data output signal.
13. A radio receiver comprising: a first signal path including a
first tuner configured to receive a first signal from a first
antenna, and a first demodulator configured to demodulate symbols
from an output of the first tuner to produce first branch metrics
derived from the demodulated symbols; a second signal path
including a second tuner configured to receive a second signal from
a second antenna, and a second demodulator configured to demodulate
symbols from an output of the second tuner to produce second branch
metrics derived from the demodulated symbols; a combiner for
maximum ratio combining the first branch metrics and the second
branch metrics; processing circuitry to process the combined first
and second branch metrics to produce an output signal; a third
signal path including a third tuner configured to receive the
second signal from the second antenna, and a third demodulator
configured to demodulate symbols from an output of the third tuner
to produce third branch metrics derived from the demodulated
symbols; a fourth signal path including a fourth tuner configured
to receive the first signal from the first antenna, and a fourth
demodulator configured to demodulate symbols from an output of the
fourth tuner to produce fourth branch metrics derived from the
demodulated symbols; a second combiner for maximum ratio combining
the third branch metrics and the forth branch metrics; and
processing circuitry to process the combined third and fourth
branch metrics to produce a data output signal.
14. The radio receiver of claim 1, wherein each of the signal paths
includes a preacquisition filter.
15. The radio receiver of claim 14, wherein each of the signal
paths includes a decimation filter preceding the preacquisition
filter.
16. A method comprising: receiving a signal on a first antenna;
producing first branch metrics derived from the signal in a first
signal path; receiving the signal on a second antenna; producing
second branch metrics derived from the signal in a second signal
path; synchronizing the first and second branch metrics are
synchronized by indexing; maximum ratio combining the first branch
metrics and the second branch metrics; and processing the combined
first and second branch metrics to produce an output signal.
17. The method of claim 16, wherein the first and second signal
paths adjust the magnitudes of the first and second branch metrics
in response to a signal-to-noise ratio of the signal in the first
and second signal paths.
18. The method of claim 16, wherein the maximum ratio combining
step sums corresponding, synchronized branch metrics from the first
and second signal paths.
19. The method of claim 16, wherein the first and second branch
metrics are synchronized prior to deinterleaving.
20. The method of claim 16, wherein the processing step is
performed by processing circuitry including a deinterleaver and a
Viterbi decoder; and like-indexed branch metrics are added when the
corresponding symbols are available from the first and second
signal paths.
21. The method of claim 16, wherein each of the signal paths
independently acquires and tracks a signal received by one of the
antennas.
22. The method of claim 21, wherein symbol and frequency tracking
for each signal path flywheels over temporary fades or outages.
23. The method of claim 16, wherein the first and second signal
paths are configured to receive an FM IBOC signal.
24. The method of claim 16, wherein when one of the signal paths
has no branch metrics available, the branch metrics of that signal
path are zeroed.
25. The method of claim 16, further comprising: using a digital
signal quality metric as a bad-track detector in at least one of
the signal paths.
26. The method of claim 25, wherein the branch metrics are zeroed
when the digital signal quality metric drops below a threshold.
27. The method of claim 26, wherein the threshold is reduced for
operation at lower signal-to-noise ratios.
28. The method of claim 16, further comprising: forcing a
reacquisition when a filtered digital signal quality metric drops
below a threshold for a predetermined number of consecutive
symbols.
29. The method of claim 16, wherein the signal paths are
independent.
30. The method of claim 16, further comprising: warping at least
one of the branch metrics at low signal-to-noise ratios to improve
maximum ratio combining performance when a signal in one or both
signal paths are degraded.
31. The method of claim 16, further comprising: using a maximum
ratio combining arbitration scheme for the two signal paths.
32. The method of claim 16, wherein the maximum ratio combining
uses shared tracking information from the two signal paths.
33. The method of claim 16, further comprising: processing the
second branch metrics to produce a data output signal.
34. A method comprising: receiving a signal on a first antenna;
producing first branch metrics derived from the signal in a first
signal path; receiving the signal on a second antenna; producing
second branch metrics derived from the signal in a second signal
path; maximum ratio combining the first branch metrics and the
second branch metrics; processing the combined first and second
branch metrics to produce an output signal; producing third branch
metrics derived from a signal in a third signal path; and
processing the third branch metrics to produce a data output
signal.
35. A method comprising: receiving a signal on a first antenna;
producing first branch metrics derived from the signal in a first
signal path; receiving the signal on a second antenna; producing
second branch metrics derived from the signal in a second signal
path; maximum ratio combining the first branch metrics and the
second branch metrics; processing the combined first and second
branch metrics to produce an output signal; producing third branch
metrics derived from a signal in a third signal path; producing
fourth branch metrics derived from a signal in a fourth signal
path; maximum ratio combining the third branch metrics and the
fourth branch metrics; and processing the combined third and fourth
branch metrics to produce a data output signal.
36. The method of claim 16, wherein symbol tracking using reference
subcarriers is started after an initial subframe is found.
37. The method of claim 36, wherein a reacquisition is invoked
within about 0.5 seconds after a digital signal quality metric if
the initial subframe is not found.
38. The method of claim 16, wherein symbol tracking is performed
using a digital signal quality metric until an initial subframe is
found.
Description
BACKGROUND
iBiquity Digital Corporation's HD Radio.TM. system is designed to
permit a smooth evolution from current analog amplitude modulation
(AM) and frequency modulation (FM) radio to a fully digital in-band
on-channel (IBOC) system. This system delivers digital audio and
data services to mobile, portable, and fixed receivers from
terrestrial transmitters in the existing medium frequency (MF) and
very high frequency (VHF) radio bands. Broadcasters may continue to
transmit analog AM and FM simultaneously with the new,
higher-quality and more robust digital signals, allowing themselves
and their listeners to convert from analog to digital radio while
maintaining their current frequency allocations. Examples of
waveforms for an FM HD Radio system are shown in U.S. Pat. No.
7,724,850, which is hereby incorporated by reference.
A variety of antenna diversity techniques have been developed and
deployed for use with automotive FM receivers. They are used to
mitigate the effects of distortion and outages due to multipath
propagation of the received FM signal, and can also accommodate the
directional characteristics of glass-embedded window antennas. All
diversity techniques use two or more antenna elements, and some
require multiple tuners/receivers. Some techniques can be applied
to digital signals, and some cannot.
Blind diversity switching can be economically attractive because a
simple multi-position switch connects the selected antenna element
to only one tuner and receiver. However, because the switching is
blind, there is no guarantee that the next antenna element will
carry a better signal, and subsequent switching may occur in rapid
succession until a good signal is found. Furthermore, since the
digital signal is coherently detected and tracked, each antenna
switching event is likely to cause symbol corruption and temporary
loss in channel state information (CSI) and coherent tracking.
Such switching transients can be avoided by using a smooth
diversity combining algorithm. These techniques involve some kind
of multiple-input signal combining (pre or post-detection), and
require multiple tuners. One combining method for analog FM signals
employs phase diversity using a constant-modulus algorithm (CMA).
However, this approach is not valid for HD Radio signals as the
digital sidebands are not characterized by a constant envelope.
IBOC HD Radio receivers can be used in combination with switch
diversity antenna systems. However the use of switch diversity
antennas introduces abrupt transients in the coherent tracking of
the digital signal, which degrades digital performance.
SUMMARY
In one aspect, the invention provides a radio receiver including a
first signal path including a first tuner configured to receive a
first signal from a first antenna, and a first demodulator
configured to demodulate symbols from an output of the first tuner
to produce first branch metrics derived from the demodulated
symbols; a second signal path including a second tuner configured
to receive a second signal from a second antenna, and a second
demodulator configured to demodulate symbols from an output of the
second tuner to produce second branch metrics derived from the
demodulated symbols; a combiner for maximum ratio combining the
first branch metrics and the second branch metrics; and processing
circuitry to process the combined first and second branch metrics
to produce an output signal.
In another aspect, a method includes receiving a signal on a first
antenna; producing first branch metrics derived from the signal in
a first signal path; receiving the signal on a second antenna;
producing second branch metrics derived from the signal in a second
signal path; maximum ratio combining the first branch metrics and
the second branch metrics; and processing the combined first and
second branch metrics to produce an output signal.
In another aspect, a method includes receiving a signal on two
antennas; demodulating the signal using two independent receiver
paths that are synchronized by symbol number; maximum ratio
combining branch metrics from the two receiver paths; and using the
combined metrics to produce an output, wherein the receiver paths
include an arbitration scheme.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a high-level Maximum Ratio Combining (MRC) block
diagram.
FIG. 2 is a functional block diagram of a receiver configured to
include FM phase diversity and digital MRC.
FIG. 3 is a functional block diagram of a receiver configured to
include either diversity (MRC and FM phase diversity) or a data
scanning receiver.
FIG. 4 is a functional block diagram of a receiver configured to
include both diversity (MRC and FM phase diversity) and a data
scanning receiver.
FIG. 5 is a functional block diagram of a receiver configured to
include both diversity (MRC and FM phase diversity) and a data
scanning receiver, also with MRC.
FIG. 6 is a functional block diagram showing a computation for
Viterbi Branch Metrics.
FIG. 7 is a functional block diagram showing a computation for
Viterbi Branch Metrics.
FIG. 8 is a graph illustrating the effects of warping of Viterbi
Branch Metrics.
FIG. 9 is a representation of branch metric scaling and
quantization.
FIG. 10 is a functional block diagram illustrating a process for
producing a digital signal quality metric.
FIG. 11 is a functional block diagram of a filter circuit.
FIG. 12 shows the spectrum of a Highpass Halfband Filter.
FIG. 13 shows the magnitude spectrum of the pre-acquisition
filter.
FIG. 14 is a functional block diagram of the quality metric
computation.
FIG. 15 is a flowchart for the acquisition of the digital
signal.
FIG. 16 shows the probability of a good acquisition.
FIG. 17 shows the probability of a bad acquisition.
FIG. 18 is a plot showing the average time required for subframe
lock.
FIG. 19 is a plot of bit error rate.
FIGS. 20 through 24 are plots of a digital signal quality
metric.
FIG. 25 is a state diagram for MRC coordination and
arbitration.
DETAILED DESCRIPTION
Maximum Ratio Combining (MRC) of Viterbi Branch Metrics (VBMs) can
afford improved signal to noise ratio (SNR) performance, at the
cost of adding a second digital reception path (from tuner to
baseband). An MRC receiver is intended to operate at lower SNRs
than a single receiver, and the acquisition and tracking algorithms
designed for a single receiver may not operate effectively at these
lower SNRs, compared to a single demodulator. Furthermore, if one
of the demodulators is outputting corrupted Viterbi Branch Metrics
(due to a poor antenna signal) while the other demodulator is
correctly demodulating, then contamination is possible, which
degrades combined performance.
Various techniques for implementing Maximum Ratio Combining in an
antenna diversity system are described herein. Such techniques are
applicable to processing of an OFDM signal of an HD Radio FM IBOC
radio system. In the embodiments described herein, MRC involves the
combining of Viterbi Branch Metrics (derived from the demodulated
symbols) from two (or possibly more) diversity receiver paths, also
referred to herein as signal paths. Each of these receiver paths
includes a tuner configured to receive a signal from a diversity
antenna element, an OFDM demodulator, and Viterbi Branch Metric
computation for each receiver output symbol (code bit). The
combining or adding of the Viterbi Branch Metrics is the MRC
function. The combined Viterbi Branch Metrics can then be
deinterleaved, decoded and processed as in the subsequent functions
of a conventional single receiver. Existing HD Radio receivers
already compute appropriate branch metrics, including signal
equalization and noise normalization, which can be used as
described herein.
Assuming independent fading at each antenna element, MRC combines
branch metrics from two different receiver paths to minimize
receiver bit error rate (BER). Branch metrics are effectively a
measure of the signal-to-noise (energy) ratio of each demodulated
symbol at the input to a Viterbi decoder. The MRC algorithm sums
corresponding, synchronized Viterbi Branch Metrics from two
receiver channels prior to deinterleaving and Viterbi decoding.
FIG. 1 is a block diagram of portions of a receiver 10 connected to
two antennas 12, 14. The receiver includes two signal paths 16, 18
(also called receiver paths or channels). The first signal path 16
includes a first radio frequency front end/tuner 20 and a first
demodulator 22. The second signal path 18 includes a second radio
frequency front end/tuner 24 and a second demodulator 26. The
antennas are configured to receive an in-band, on-channel (IBOC)
radio signal, which can be an FM HD Radio signal. HD Radio signals
are described in, for example, U.S. Pat. No. 7,933,368, which is
hereby incorporated by reference. Each signal path includes
processing circuitry or a processor programmed to compute Viterbi
Branch Metrics for each receiver output symbol. In the example of
FIG. 1, such processing circuitry or a processor can be included in
the demodulator blocks.
The antennas can be elements with different characteristics,
positioned at different locations, and/or positioned at different
orientations. The demodulators produce Viterbi Branch Metrics on
lines 28 and 30. These Viterbi Branch Metrics are maximum ratio
combined in combiner 32. The combined metrics are then passed to
circuitry 34 that processes the combined metrics to produce an
output signal on line 36. This processing circuitry may include a
deinterleaver, decoder, codec, etc. as is known in the art.
MRC is attained by adding the corresponding Viterbi Branch Metrics
of the demodulated symbols (bits before decoding) from the two
receiver paths. Corresponding VBMs from the two receiver paths can
be synchronized, as shown by line 38, by indexing prior to the
deinterleaver. The indexing is used to unambiguously identify and
label (number) the symbols in the interleaver matrix. Like-indexed
VBMs are added when the corresponding symbols are available from
both receiver paths. The embodiment of FIG. 1 uses two independent
receiver/demods, then identifies and combines like-indexed symbol
Viterbi Branch Metrics. The indexing allows the two receiver paths
to operate asynchronously.
When one of the receiver paths has no VBMs available, the missing
VBMs are assumed to be zero, and only the receiver path with valid
VBMs is used (no addition is necessary). When VBMs are not
available from either receiver, the downstream functions
(deinterleaver, Viterbi decoder, etc.) are reset, assuming for
example, a reacquisition process is invoked.
A baseline MRC technique assumes that each receiver path is
configured to independently acquire and track the signal, and that
the branch metrics are aligned and combined. In one case,
performance is near optimum when both receivers are tracking the
signal with proper synchronization. The main performance
enhancement is achieved in dynamic fading conditions. When one
antenna is in a deep fade, the other antenna may not be faded, and
vice versa. Symbol and frequency tracking for each receiver path
can flywheel over short fades or outages. The flywheeling maintains
adequate synchronization during brief signal outages.
Analog FM phase diversity can be implemented using two antennas,
two tuners, and two FM receiver paths. A pair of signals can be
combined prior to the FM demodulator using a Constant Modulus
Algorithm (CMA), or some variation thereof. Since two antenna
signal paths are available, these phase diversity systems are
compatible with MRC for IBOC digital diversity.
FIG. 2 shows a functional block diagram of an implementation of
digital MRC in a vehicle application that also employs analog FM
phase diversity. FIG. 2 is a block diagram of a receiver 40
connected to two antennas 42, 44. The receiver includes two signal
paths 46, 48. The first signal path 46 includes a first radio
frequency front end/tuner 50, an analog FM demodulator 52 and a
first digital demodulator 54. The second signal path 48 includes a
second radio frequency front end/tuner 56, the analog FM
demodulator 52 and a second digital demodulator 58. The antennas
are configured to receive an in-band, on-channel radio signal,
which can be an FM HD Radio signal. The antennas can be elements
with different characteristics, positioned at different locations,
and/or positioned at different orientations. The demodulators
produce Viterbi Branch Metrics on lines 60 and 62. These VBMs are
maximum ratio combined in combiner 64. The combined metrics are
then passed to circuitry 66 that processes the combined metrics.
This processing circuitry may include a deinterleaver, decoder,
codec, etc. as is known in the art. An audio decoder 68 produces
digital audio and blend control signals as illustrated by line 70.
The analog FM demodulator 52 includes FM diversity processing 72
and FM demodulation 74 to produce a demodulated FM signal on line
76. A blend control 78 blends the demodulated FM signal on line 76
and the digital audio signal to produce an audio output on line 80.
Each receiver path is configured to calculate the branch metrics
and to independently acquire and track the signal, and ensure that
the branch metrics are aligned and combined.
MRC is attained by adding the corresponding VBMs of the demodulated
symbols (bits before decoding) from the two receiver paths.
Corresponding VBMs from the two receiver paths can be synchronized,
as shown by line 82, by indexing prior to the deinterleaver.
FIGS. 3 through 5 show several implementation options for including
MRC in a data scanning receiver. FIG. 3 shows how a receiver 90
with two antenna signal paths can be configured to use the second
antenna signal path for either MRC and phase diversity, or for a
non-MRC scanning data channel, but not both simultaneously. The
receiver 90 is connected to two antennas 92, 94. The receiver
includes two signal paths 96, 98. The first signal path 96 includes
a first radio frequency front end/tuner 100 that can be tuned to a
first frequency, and a first digital demodulator 102. The second
signal path 98 includes a second radio frequency front end/tuner
104 that can be tuned to either the first frequency or a second
frequency, and a second digital demodulator 106. The antennas are
configured to receive an in-band, on-channel radio signal, which
can be an FM HD Radio signal. The antennas can be elements with
different characteristics, positioned at different locations,
and/or positioned at different orientations. The demodulators
produce Viterbi Branch Metrics on lines 108 and 110. These VBMs can
be maximum ratio combined in combiner 112. The combined metrics are
then passed to circuitry 114 that processes the combined metrics to
produce an output signal. This processing circuitry may include a
deinterleaver, decoder, codec, etc. as is known in the art.
Alternatively, instead of MRC, additional processing circuitry 116
can be provided to process the output from the second digital
demodulator to produce a data output on line 118. The tuner outputs
are subject to FM diversity processing and/or analog FM
demodulation as shown in block 120 to produce an analog audio
signal on line 122. The analog FM audio signal and a digital audio
signal on line 124 are blended as shown in block 126 to produce an
audio output on line 128. Each receiver path is configured to
calculate the branch metrics and to independently acquire and track
the signal, and ensure that the branch metrics are aligned and
combined.
FIG. 4 is a block diagram of a receiver 130 that includes many of
the elements of FIG. 3 and adds a third signal path 131. The third
signal path includes a third tuner 132 and a third digital
demodulator 133 to enable both MRC and phase diversity as well as a
non-MRC data scanning channel. The output of the third digital
demodulator is processed by processing circuitry 134 to produce a
data output on line 135. In this example, two of the three tuners
are tuned to the same frequency.
FIG. 5 is a block diagram of a receiver 136 that includes many of
the elements of FIG. 4 and adds a fourth signal path 137. The
fourth signal path includes a fourth tuner 138 and a fourth digital
demodulator 139 to enable MRC on both the main receiver signal as
well as the scanning data path. The Viterbi branch metric outputs
of the third and fourth digital demodulators on lines 140 and 141
are combined in combiner 142. Then the combined signal is processed
by processing circuitry 143 to produce a data output on line 144.
In this example, both the first and second tuners are tuned to a
first frequency, and both the third and fourth tuners are tuned to
a second frequency.
I. Viterbi Branch Metrics
The Viterbi Branch Metrics (VBMs) for the described IBOC MRC
embodiments are a ratio of the estimated signal to noise energy of
the channel symbols (bits) prior to deinterleaving and decoding.
These VBMs can be computed as described in U.S. Pat. Nos.
6,982,948, 7,305,056 or 7,724,850, which are hereby incorporated by
reference. The first two patents (U.S. Pat. Nos. 6,982,948 and
7,305,056) use linear filters to estimate Channel State Information
(CSI).
As shown in U.S. Pat. No. 7,305,056, an HD Radio signal includes an
analog modulated carrier and a plurality of digitally modulated
subcarriers. Some of the digitally modulated subcarriers are
reference subcarriers. FIG. 6 is a functional block diagram
describing the CSI estimation using linear filters as shown in U.S.
Pat. No. 7,305,056. FIG. 6 illustrates a method of estimating both
the phase reference and the CSI from the reference subcarriers in
an HD Radio signal. The reference subcarriers can be used for
acquisition, tracking, estimation of CSI and coherent
operation.
As shown in FIG. 6, the complex training symbols carried by the
reference subcarriers are input on line 148 and the complex
conjugate of the symbols is taken as shown in block 150. The
complex conjugate is multiplied with a known training sequence on
line 152 by multiplier 154. This removes the binary (.+-.1) timing
sequence modulation from the received training subcarriers by
multiplying them by the synchronized, decoded, and
differentially-reencoded BPSK timing sequence. The resulting
symbols on line 156 are processed by a finite impulse response
(FIR) filter 158 to smooth the resulting symbols over time,
yielding a complex conjugated estimate of the local phase and
amplitude on line 160. This value is delayed by time delay 162 and
multiplied by an estimate of the reciprocal of the noise variance
on line 164 by multiplier 166. The noise variance is estimated by
subtracting the smoothed estimate of the local phase and amplitude
on line 160 from the input symbols (after appropriate time
alignment provided by delay 168) at summation point 170. Then
squaring the result as shown in block 172, and filtering the
complex noise samples as illustrated in block 174. The reciprocal
is approximated (with divide-by-zero protection) as shown in block
176. This CSI weight is interpolated over the 18 subcarriers
between pairs of adjacent training subcarriers as illustrated by
block 178 to produce resulting local CSI weights on line 180. The
CSI weights are then used to multiply the corresponding local
data-bearing symbols received on line 182, after they have been
appropriately delayed as shown in block 184. Multiplier 186 then
produces the soft decision output on line 188.
In FIG. 6, lines carrying training symbols are labeled T and lines
carrying data are labeled D. In addition, filter 174 includes a
delay of:
.gtoreq..beta..times..times..beta. ##EQU00001## and, y.sub.n
m=2(1-.beta.)y.sub.n-1,m-(1-.beta.).sup.2y.sub.n-2,m+.beta..sup.2x.sub.n,-
m.
These expressions relate to a 2-pole IIR filter with a time
constant .beta.. The IIR filter computes smoothed output samples
"y" from input sample "x" and previous output samples.
The CSI weight combines the amplitude weighting for maximum ratio
combining along with a phase correction for channel phase errors.
This CSI weight is dynamic over time and frequency, and is
estimated for each QPSK symbol.
.times..times..times..times..times..times..alpha..sigma.
##EQU00002## where .alpha.* is an estimate of the complex conjugate
of the channel gain and .sigma..sup.2 is an estimate of the
variance of the noise.
The operation of the CSI recovery technique of FIG. 6 assumes
acquisition and tracking of the frequency of the subcarriers, and
the symbol timing of the OFDM symbols. The frequency and symbol
timing acquisition techniques exploit properties of the cyclic
prefix. The frequency and symbol tracking is accomplished through
observation of the phase drift from symbol to symbol over time or
frequency (across subcarriers).
After acquisition of both frequency and symbol timing,
synchronization to the Block Sync pattern of the BPSK Timing
Sequence is attempted by cross-correlating the differentially
detected BPSK sequence with the Block Sync pattern. The
differential detection is performed over all subcarriers assuming
that the location of the training subcarriers is initially unknown.
A cross-correlation of the known Block Sync pattern with the
detected bits of each subcarrier is performed. A subcarrier
correlation is declared when a match of all 11 bits of the Block
Sync pattern is detected. Block synchronization (and subcarrier
ambiguity resolution) is established when the number of subcarrier
correlations meets or exceeds the threshold criteria (e.g., 4
subcarrier correlations spaced a multiple of 19 subcarriers
apart).
After Block Sync is established the variable fields in the BPSK
Timing Sequence can be decoded. The differentially detected bits of
these variable fields are decided on a "majority vote" basis across
the training subcarriers such that decoding is possible when some
of these subcarriers or bits are corrupted. The 16 Blocks within
each Modem Frame are numbered sequentially from 0 to 15. Then the
most significant bit (MSB) of the Block Count field is always set
to zero since the Block Count never exceeds 15. Modem Frame
synchronization is established with knowledge of the Block Count
field.
The coherent detection of this signal requires a coherent phase
reference. The decoded information from the BPSK Timing Sequence is
used to remove the modulation from the training subcarriers leaving
information about the local phase reference and noise. Referring to
FIG. 6, the binary (.+-.1) timing sequence modulation is first
removed from the received training subcarriers by multiplying them
by the synchronized, decoded, and differentially-reencoded BPSK
Timing Sequence. A FIR filter is used to smooth the resulting
symbols over time, yielding a complex conjugated estimate of the
local phase and amplitude. This value is delayed and multiplied by
an estimate of the reciprocal of the noise variance. The noise
variance is estimated by subtracting the smoothed estimate of the
local phase and amplitude from the input symbols (after appropriate
time alignment), squaring and filtering the complex noise samples,
then approximating the reciprocal (with divide-by-zero protection).
This CSI weight is interpolated over the 18 subcarriers between
pairs of adjacent training subcarriers. The resulting local CSI
weights are then used to multiply the corresponding local
data-bearing symbols.
In one embodiment, the low pass filter 158 in FIG. 6 is an 11-tap
FIR filter. The 11-tap FIR filter is used to dynamically estimate
the complex coherent reference gain .alpha. at each reference
subcarrier location for each symbol time. The filtering over time
with the 11-tap FIR filter, and subsequent filtering across
subcarriers is performed to compute a local estimate of the
coherent reference gain .alpha. for each QPSK symbol location over
both time and frequency. A larger FIR filter with more taps would
reduce the estimation error when the signal statistics are
stationary, but the bandwidth would be too small to track
Doppler-induced changes in the signal at maximum highway speeds.
Therefore 11 taps with a tapered symmetric Gaussian-like impulse
response was considered to be appropriate. A symmetric FIR is used
instead of an IIR filter for its linear phase property which has
zero bias error for a piecewise linear (approximately) channel
fading characteristic over the span of the filter. This smoothed
coherent reference signal output of the FIR filter is subtracted
from the delayed input samples to yield the instantaneous noise
samples. These noise samples are squared and processed by an IIR
filter 174 to yield an estimate of the noise variance
.sigma..sup.2. This filter has a narrower bandwidth than the FIR
filter to yield a generally more accurate estimate of the noise
variance. After appropriate sample delays to match the filter
delays, the symbol weight .alpha.*/.sigma..sup.2 is computed for
each subcarrier. These values are smoothed and interpolated across
the subcarriers for each OFDM symbol to yield more accurate
estimates. This weight is unique for each OFDM symbol and each
subcarrier providing a local (time and frequency) estimate and
weight for the symbols forming the branch metrics for a subsequent
Viterbi decoder.
As used herein, the "complex coherent reference gain (.alpha.)" of
a QPSK symbol (depending on time/frequency location since it is
dynamic) is defined as .alpha.. It is a complex term, including
real and imaginary components, that represents the gain and phase
of the symbol associated with it. This value is estimated by the
processing and filtering described. The "composite coherent channel
reference signal x.sub.n" is the composite value of .alpha.
computed over all the reference subcarriers over any one OFDM
symbol time.
The multiple roles of the Reference Subcarriers for acquisition,
tracking, estimation of channel state information (CSI) and
coherent operation have been described in the incorporated patents.
The system of U.S. Pat. No. 7,305,056 was designed to accommodate
vehicles with fixed antennas. The system was designed for coherent
operation in the FM broadcast band (88-108 MHz) with fading
bandwidth to accommodate vehicles at highway speeds. The various
coherent tracking parameters are estimated using filters with
bandwidths that approximate the maximum expected Doppler bandwidth
(roughly 13 Hz). With a fixed antenna, the pertinent tracking
statistics of the input signal to the tracking algorithms are
assumed to vary at a rate no greater than the Doppler
bandwidth.
As used herein, the "complex coherent reference gain (.alpha.)" of
a QPSK symbol (depending on time/frequency location since it is
dynamic) is defined as .alpha.. It is a complex term, including
real and imaginary components, that represents the gain and phase
of the symbol associated with it. This value is estimated by the
processing and filtering described. The "composite coherent channel
reference signal x.sub.n" is the composite value of .alpha.
computed over all the reference subcarriers over any one OFDM
symbol time.
The third patent (U.S. Pat. No. 7,724,850) uses nonlinear filters
to estimate CSI. The nonlinear filters improve performance in the
presence of impulse noise and step transients. Step transients can
be caused by stepped age or by switching antenna diversity systems.
This patent is listed below, and a functional block diagram is
shown in FIG. 7.
FIG. 7 shows an example wherein the 11-tap FIR filter is replaced
with a 5-tap median filter. The goal of the process(es) shown here
is to provide estimates of the coherent channel complex gain ("a"
values) along with estimates of the noise or interference. These
estimates are local in time and frequency (subcarrier location) to
accommodate the dynamic selective fading channel experience in a
mobile environment such as a moving automobile. These estimates are
derived from the reference subcarrier symbols which have been
stripped from the received and demodulated signal as previously
described, and are input on line 250 as S.sub.r,n complex values.
The data used to modulate these symbols is already known and
removed from these symbols with the first conjugate multiply
operation (illustrated by multiplier 252) to yield the
instantaneous complex channel gain values a2.sub.r,n on line 254.
The subsequent median filtering 256 in time reduces the noise while
maintaining the step changes due to antenna switching to produce
intermediate values a1.sub.r,n on line 258. These intermediate
values are further filtered (smoothed) over the reference
subcarriers (in frequency) as shown in block 260 to produce the
final complex channel gain values a.sub.r,n. These a.sub.r,n, gain
values are later used outside this algorithm to process (equalize
and provide branch metric information) the signal constellations
for the data bearing symbols in the conventional manner for QAM
symbol demodulation.
The next step in this process is to estimate the noise associated
with each of these complex channel gain values. The instantaneous
noise samples are estimated by subtracting the a.sub.r,n-2 values
from the (appropriately delayed) noisy corresponding input samples
a2.sub.r,n-2, as illustrated by summation point 262. As shown in
block 264, the magnitude-squared value is computed from these
complex noise samples to yield the instantaneous noise variance
estimates var.sub.n-2 on line 266. These instantaneous noise
variance samples are poor estimates of the local (time and
frequency) noise and require processing and filtering to produce
useful noise variance estimates. Although simpler time and
frequency filtering would normally be used to reduce the error of
these instantaneous noise variance estimates, this type of
filtering would not effectively accommodate the changing noise due
to fading, Automatic Gain Control AGC action and step changes due
to antenna switching. Therefore a median filter 268 is used to
filter these instantaneous variance samples in time to produce
samples varflt.sub.n-16 and conventional (linear IIR or FIR filter
270) filtering is used to further smooth across frequency
(subcarriers) to produce the final variance estimates
.sigma..sup.2.sub.r,n-16 in a manner similar to the complex channel
gain estimates above. An additional feed forward path 272 is
provided to capture the relatively large noise impulses that occur
due to the antenna switching. When these values (scaled by a factor
0.5 as shown in block 274) exceed the median-filtered estimate,
then these larger values are selected for output to the frequency
smoothing filter by the select max function illustrated in block
276. These values are then smoothed over the reference subcarriers
as shown in block 278. This is important in subsequent formation of
the branch metrics which exploits this knowledge of the large noise
impulses.
Analyses and simulation of the algorithm improvements to the
coherent reference estimation just described appear to work
sufficiently well for the cases analyzed and simulated. These cases
include a flat and selective fading channel with Doppler bandwidth
consistent with highway speeds and noise as low as 0 dB SNR.
However other channel conditions should be considered, such as
impulsive noise, or residual transient effects not entirely
suppressed by the new coherent reference processing. In this case
the adjusted coherent reference values of x are appropriate;
however, the noise variance estimate would be corrupted. The noise
impulse could be high for the symbol(s) where the impulse occurred,
but the IIR filter would suppress this noise estimate value at the
impulse instant, and spread the noise estimate over the impulse
response time of the IIR filter. It would be preferable in this
case to feed-forward the high noise samples in parallel with the
IIR path (with appropriate delay matching). For symbols where the
noise pulse is sufficiently higher than the IIR filter output, this
noise pulse should be used to determine the estimated noise
variance for those symbols. When the feed-forward path is used for
these noise pulses, the energy into the IIR filter for these
samples should be reduced so that the local noise peak is not
spread over the span of the IIR filter. It is easy to consider
several variations of this process for handling noise peaks in the
noise variance estimate.
The noise variance estimation process is modified to improve
performance with switching transients and to accommodate a faster
AGC. The original noise estimation employed a 2-pole IIR filter
with parameter a= 1/16 (not to be confused with the subscripted
"a.sub.r,n," value notations for the complex channel gains). The
peak of the impulse response of this filter was at a delay of 8
samples (symbols), although the decaying tail was much longer
making the step delay closer to 16 samples (symbols).
The functions described in FIGS. 6 and 7 can be performed, for
example, in the digital demodulator blocks of FIGS. 1-5.
According to embodiments of the invention, these branch metrics can
be modified as described below in order to optimize their use in
maximum ratio combining for an FM IBOC diversity system, including
adjusting for non-linear filtering effects, warping, quantization,
and synchronization, as described in the following sections.
Analysis of Viterbi Branch Metrics
The relationship between carrier-to-noise ratio Cd/No and the VBM
values is analyzed in this section, as this relationship influences
the modifications described in subsequent sections. The VBMs are
formed by multiplying the received symbols by the computed
CSIweight. These channel state information (CSI) weights are
derived from the reference subcarriers and interpolated over the 18
data-bearing subcarriers between the neighboring pairs of reference
subcarriers. This CSIweight combines the amplitude weighting for
MRC along with a phase correction for channel phase errors.
.times..times..times..times..times..times..alpha..sigma.
##EQU00003##
where a* is the complex conjugate of the estimated channel gain,
relative to a quadrature phase shift keying (QPSK) symbol energy of
one, and .sigma..sup.2 is an estimate of the variance of the noise
for a QPSK symbol. Since the noise variance is estimated in two
dimensions for the QPSK symbol, then .sigma..sup.2=No (instead of
.sigma..sup.2=No/2 usually associated with a one-dimensional
matched filter). The QPSK symbol has a nominal magnitude of |a|=
{square root over (Es)}= {square root over (2Ec)}, where Es is the
energy of a QPSK symbol, and Ec is the energy of one of the two
code bits of the QPSK symbol. When a received bit is multiplied by
the CSI weight, it has a typical (absolute) value of
.sigma. ##EQU00004##
The code bit energy Ec is expressed as a function of the total
digital signal power Cd.
##EQU00005##
Then the typical (absolute) value of the branch metrics can be
expressed as a function of Cd/No.
.times..times..times..times..times..times. ##EQU00006##
Adjustments For Nonlinear Filtering
The branch metric analysis described above assumes ideal linear
filtering. However, current HD Radio receiver implementations
employ several nonlinear filtering techniques to mitigate the
undesirable effects of impulsive noise and step transients due to
automatic gain control (AGC) and/or switched diversity antenna
systems, as is described in U.S. Pat. No. 7,724,850, which is
hereby incorporated by reference. The branch metric relationship
with Cd/No can be adjusted to allow for gain difference with these
nonlinear filters. As shown above, the typical branch metric
relationship for the ideal linear filter model is
|VBM|=Cd/No-52.7dB
A functional block diagram of the CSI estimate technique using
nonlinear filtering is shown in FIG. 7. The top signal path in FIG.
7 shows a 5-tap median filter. This filters the complex QPSK
(constrained to BPSK) symbols of the reference subcarriers that
have been stripped of data. The symbol values represent the complex
channel gain for each reference subcarrier. The median filter in
this case does not impose a bias relative to the weighted complex
sample mean that would be obtained by linear filtering in the case
of all-white Gaussian noise (AWGN). This is because the
two-dimensional Gaussian noise probability density function is
symmetric about the mean complex value of the QPSK symbol.
The 7-tap median filter for the noise variance estimate produces a
bias relative to a linear averaging filter. This is because the
squared error samples have a nonsymmetric distribution about the
mean. Specifically the sum of the square of the pair of unit
variance Gaussian samples produces a Chi-squared (.sigma..sup.2)
distribution with 2 degrees of freedom, having a mean of 2 (2
dimensions) and a distribution of
.function.e.function.e ##EQU00007##
The variance of the noise is the mean of the Chi-squared
distribution.
.sigma..intg..infin..times..function..times.d ##EQU00008##
The nonlinear filter in the receiver implementation approximates
the variance with the median of the Chi-squared distribution, and
is solved by
.intg..times..function..times.d.intg..infin..times..function..times.d.fun-
ction..apprxeq. ##EQU00009##
The median value of 1.386 is relative to a linear mean of 2,
yielding a gain of ln(2)=0.693 instead of unity gain expected of a
linear filter. However, the median of a finite number of samples
(e.g., 7) is biased slightly higher than the true median of a large
sample set. A simple simulation of a sliding 7-tap median filter
over 1 million Chi-squared samples reveals that the gain is
approximately 0.76 (instead of unity gain for linear filters),
underestimating the noise variance by 1.2 dB. This is due to the
asymmetry of the distribution of the square of the Gaussian complex
samples. Then this will tend to overestimate the CSIweight by a
factor of about 1.316 (1.2 dB).
There is another filter nonlinearity due to the excess short term
noise estimates. In this case large impulsive noise samples (scaled
by 0.5) will be selected as the noise-squared filter output. The
result is that the feedforward peak excess short term noise
estimates will overestimate the noise. The net result of both
nonlinearities (7-tap median filter, and select max) is that the
noise variance is underestimated by a factor of 0.83 (0.8 dB), so
the CSIweight is overestimated by a factor of 1.2.
Simulation results of an actual receiver show the mean branch
metric values as a function of Cd/No. The results show that the
simulated branch metrics are a factor of 1.073 (0.3 dB error)
greater than the predicted values at typical Cd/No operating
points, even after correction for nonlinear filtering. One
explanation why the VBMs are larger than predicted is that the
finite symbol estimation (e.g., 5-tap median filter) is influenced
by the nonzero median of the noise over those 5 samples. The symbol
magnitude would be overestimated (although not biased) at the
median filter output because of the vector addition of the noise
component. This would also result in underestimation of the noise
variance because the symbol median is subtracted from the other
samples, then squared to produce noise energy samples. The net
error would be difficult to analyze because of the complication of
additional filtering across reference subcarriers. However, this
small error is assumed acceptable as sufficient verification of the
filter gain for analysis in subsequent sections.
For these reasons and according to embodiments of the invention
described herein, the computed branch metric prediction for
nonlinear filtering should include an overall adjustment of about
2.3 dB (1.2+0.8+0.3 dB). |VBM|=Cd/No-52.7+2.3=Cd/No-50.4dB
Branch Metric Warping
The ideal branch metrics increase in proportion to Cd/No. However,
at low SNR, the channel symbols become overestimated. For example,
the channel symbols estimated by the 5-tap median filter will
generally have a non-zero median even when no signal is present.
That is because the channel symbol is the median of the 5 noise
samples. This will cause an underestimation of the noise variance.
So, branch metrics are overestimated (warped) at low SNR. The
expression for the CSIweight can be modified to "unwarp" the values
at low SNR. This can be accomplished by multiplying the existing
CSIweight with a warp factor CSIwarp. CSIweightw=CSIweightCSIwarp
where
.sigma..times..times..times..times..sigma. ##EQU00010##
The value of parameters c and p can be empirically adjusted for
best performance. The value of Cd/No is related to the nominal
branch metric magnitude, including the gain correction factor for
the nonlinear filtering.
.function..sigma..times..times. ##EQU00011## .function..sigma.
##EQU00011.2##
The plots of FIG. 8 show the effects of CSIwarp over a range of
Cd/No, that is, the suppression of branch metrics at low SNR.
Simulation results suggest using c=0.25, p=2, since it tends to
offer the best performance over various conditions. As used in this
description, "low SNR" means near zero dB (Ec/No) or lower.
Branch Metric Quantization
Memory constraints are satisfied by imposing quantization on the
branch metrics. Quantization is determined by the number of bits
used to represent the VBMs. Although 8 bits of quantization have
been used, it is desirable to reduce this to fewer (e.g., 4) bits.
The optimal quantization zone width (quantization resolution) is
defined by the following formula:
##EQU00012## where No is a noise power spectral density, b is the
number of bits for a soft decision, and T is in units of {square
root over (Ec)}. So at Ec/No=1, the quantized value of the branch
metric should be {square root over (2.sup.b)}. The computed VBM in
an IBOC receiver already has a factor of {square root over (2)} in
the computation, as well as a factor of about 1.3 due to the
nonlinear filtering gain.
.sigma. ##EQU00013##
Then the practical scale factor for the IBOC receiver branch
metrics should be: scale=0.544 {square root over (2.sup.b)}
Branchmetric.sub.--nzq=max{-2.sup.b-1+1,min{2.sup.b-1-1,round[scaleBranch-
metric.sub.--nz]}}.
In one example, for b=4 bits of quantization, the scale factor
could be scale=2.17. So .+-.4 would represent the quantized values
at Ec/No=1, about Cd/No=54.2 dB_Hz. The maximum range is +7/-8,
about 3 dB greater than .+-.Ec/No.
FIG. 9 is a diagram showing scaling (scale=2.17) and quantization
for Viterbi Branch Metrics. In FIG. 9, the numbers are the actual
integer quanta values represented with 4 bits (16 possible numbers
in 2's complement). For this example, an integer value of 4 (or
.+-.4) is equivalent to Ec/No=1, or zero dB, where Ec/No is the
code bit energy divided by the noise density.
The combined effects of scaling, quantization and warping were
simulated to empirically determine the parameter settings for
warping (p and c) as well as the scale factor associated with VBM
quantization bits. These simulation results suggest a different
scale factor than the previous analysis. Table 1 shows the
recommended scale values for various quantization choices (bits for
VBM).
The benefits of warping are best measured with one sideband, since
the warping mitigates contamination from the missing sideband due
to nonzero (noisy) VBMs. VBM quantization with best scaling was
simulated (except an additional VBM scale factor of 32 was also
used for 8-bit quantization to ensure saturation in the case of
high impulsive noise samples). The recommended warping parameters
are c=0.25, p=2. Over all conditions simulated, for 4-bit
quantization, the loss is less than half a dB. For 3-bit
quantization, it is less than one dB (with warping). For 2-bit
quantization, degradation is less than 2 dB (again, with warping).
The best choices for scale factor for each VBM quantization (bits)
are bolded in Table 1.
TABLE-US-00001 TABLE 1 VBM Quantization Loss with Measured Best
Scaling, BER Results of Matlab FM Simulation Warping, c = 0.25, p =
2, Seed = 100, 5/24-25/12 BER Degrad. (dB) vs. BER Degrad. (dB) vs.
VBM Floating-Point VBMs, BER Degrad. (dB) vs. Floating-Point VBMs,
Quantization VBM AWGN @ 56 dB-Hz with Floating-Point VBMs, UF
Rayleigh Fading @ (bits) Scaling Warping One Sideband Disabled AWGN
@ 54 dB-Hz 57 dB-Hz Float NA OFF -- -- -- Float NA ON -0.3 0.05
0.05 8 32 OFF 0.05 0.01 0 8 32 ON -0.05 0.08 0.12 8 8.704 OFF -0.15
0.06 0.04 8 8.704 ON -0.3 0.07 0.12 4 3.5 OFF 0.48 0.07 0.33 4 3.5
ON 0.17 0.11 0.34 3 2.5 OFF 1.11 0.3 0.71 3 2.5 ON 0.76 0.31 0.79 2
2 OFF 2.24 0.82 1.75 2 2 ON 1.22 0.85 1.68
Synchronization of VBMs
In the disclosed embodiments, both of the first and second receiver
signal paths may operate independently (asynchronously). The VBMs
from each receiver path are combined when available. Both receiver
paths use their own acquisition and tracking, and the branch
metrics must be aligned for combining. When only one receiver path
has valid branch metrics, then the branch metrics from the other
receiver path are not added.
Performance is near optimum when both receiver paths are tracking
the signal with proper synchronization. The main performance
enhancement is achieved in dynamic fading conditions. When one
antenna is in a deep fade, the other antenna may not be faded, and
vice versa. Tracking can flywheel over short fades or outages.
When one of the receiver paths is not tracking the signal, its
branch metrics are effectively zero and MRC offers no additional
advantage to the tracking demodulator, except to improve the
probability that at least one demodulator is decoding the signal.
This situation could be improved if tracking information were
shared between receivers. The loss may be apparent in AWGN where
tracking can be lost due to operation below the single-receiver SNR
threshold, where the combining gain would offer sufficient bit
error rate (BER) performance if this receiver was tracking.
Alternatively, both receiver paths could share synchronization
based on both antenna signals. This option offers better
performance, but extensive demodulator software modifications are
required over the single demodulator. Alignment between branch
metrics is trivial because both receiver paths are already
synchronized. The acquisition and tracking is common to both signal
paths. Synchronization between the pair of input signal paths
should be ensured, and tuner local oscillator frequencies should be
locked. Performance is improved under all conditions. The
acquisition and tracking performance is improved along with the
signal decoding BER performance.
II. Acquisition and Frame Synchronization Using DSQM
As previously stated, an FM IBOC receiver that implements MRC
operates at low SNR conditions. Existing IBOC receivers use
parameters for acquisition and frame synchronization that can,
according to embodiments of the invention described herein, be
optimized for these low SNR conditions using a Digital Signal
Quality Metric (DSQM).
The Digital Signal Quality Metric (DSQM) is an algorithmic function
used to measure (compute) the quality of a digital OFDM signal. The
DSQM is a number ranging from zero to one, indicating the viability
of the digital sidebands of an FM IBOC signal. A value near zero
indicates that no useful signal is detected, while a value near one
indicates that the signal quality is nearly ideal. A midrange value
of 0.5, for example, indicates a corrupted but possibly decodable
digital signal. U.S. Pat. No. 7,933,368 describes the DSQM
function, and is hereby incorporated by reference.
The DSQM has several applications: 1) detect a viable digital
signal channel for digital seek/scan, 2) establish initial symbol
synchronization and carrier frequency offset for digital signal
acquisition, 3) assess antenna element signal quality for diversity
switching and MRC, where a more-efficient version, DSQM-lite,
exploits knowledge of existing symbol synchronization.
FIG. 10 is a functional block diagram of DSQM processing. Upper
sideband and lower sideband signals are received on lines 300 and
302 respectively. These signals can be received from sideband
isolation filters at 186 ksps (where decimation by 2 filters are
used for 372 ksps). The signals are shifted to base band in mixers
304 and 306. Preacquisition filters 308 and 310 filter the baseband
signals. Signal quality metrics Q and peak index P for each digital
sideband are determined as shown in blocks 312 and 314. Then the
combined quality metrics Q and peak index P are used to compute a
DSQM as shown in block 316. An estimate of symbol timing and
(sub)carrier frequency offset for the initial acquisition case is
computed as shown in block 318.
The DSQM computation shown in FIG. 10 is comprised of 5 related
components: 1) shift center frequency of the preacquisition signal
bandwidth to baseband, 2) preacquisition filter each sideband, 3)
compute signal quality metrics Q and peak index P for each digital
sideband, 4) combine the signal quality metrics to form composite
DSQM, and 5) estimate symbol timing and (sub)carrier frequency
offset for the initial acquisition case.
A portion of the USB and LSB signal bandwidths is used for the DSQM
estimation. In one example, the desirable frequency portion is
centered at about 155 kHz for the USB, and -155 kHz for the LSB. A
bandwidth of about 46.5 kHz is useful for DSQM because it allows
for suppression of a potential first-adjacent analog signal.
Nyquist sampling of these signals results in efficient
computation.
The DSQM also estimates receiver symbol boundary and frequency
error caused by different transmitter and receiver reference
oscillators and symbol boundary uncertainty. Its one-time
corrections are applied prior to the start of demodulation;
synchronization is maintained thereafter by tracking control in the
demodulator.
A more detailed description of DSQM is provided below, in Section
III. DSQM Algorithm Description. A more efficient implementation of
DSQM, called DSQM-lite, can be used for antenna diversity
switching. U.S. patent application Ser. No. 13/165,325, filed Jun.
21, 2011 and titled "Method And Apparatus For Implementing Signal
Quality Metrics And Antenna Diversity Switching Control", describes
the DSQM-lite function, and is hereby incorporated by reference.
The efficiency of DSQM-lite is derived from knowledge of the symbol
synchronization after the signal has been acquired. Instead of
processing the entire symbol vector, the DSQM is computed only for
the synchronized samples within the symbol.
DSQM and/or DSQM-lite can be used to optimize parameters for the
use of MRC antenna diversity in a receiver operating at a lower
SNR. Since performance in AWGN can improve as much as 3 dB, the
acquisition and tracking should also be capable of operating 3 dB
lower. Even greater improvements in reception sensitivity are
possible in fading. However, the fourth-power symbol tracking in
the previously used demodulator implementations breaks down at
these lower SNR operating conditions. Thresholds on DSQM and
correlation requirements with sync patterns in the reference
subcarriers for frame sync could be modified to improve acquisition
at lower SNR. A functional block diagram of a receiver employing
the MRC antenna diversity technique for OFDM signals is shown in
FIG. 1.
A signal processing strategy described below includes eliminating
the fourth-power symbol tracking, along with the fourth-power
"Badtrack" detection. Badtrack is a condition where the symbol
tracking settles somewhere other than the actual symbol boundary,
and remains stuck there. The fourth-power technique is commonly
used to strip the data phase modulation imposed upon QPSK symbols.
This leaves the complex gain information used to estimate Channel
State Information (CSI) that is used in subsequent Viterbi Branch
Metric computations. The fourth-power operation multiplies the
angle of the complex gain by 4. This is remedied by dividing the
resulting angle by 4 to yield the channel phase. However, it also
multiplies the noise by a factor of 4. This is typically acceptable
for operation of a single receiver, since the acquisition and
tracking algorithm based on the fourth-power operate acceptably at
the lowest SNR for useful data. However, the lower operating SNR of
an MRC receiver is prevented because of the increased noise due to
fourth-power processing, so an alternate technique is sought.
Since the fourth-power processing is discarded, the symbol tracking
is left to flywheel using the symbol timing sample offset
determined by DSQM during this period. The sample timing error will
drift during this time due to clock error (e.g., 100 ppm results in
18.6 samples/sec drift at 186 kHz sample rate). If the symbol
timing drifts too far, then the symbol tracking loop may not be
able to converge to the correct operating point. The symbol
tracking using reference subcarriers is started after an initial
subframe is found, which will prevent further symbol timing drift.
Therefore, to avoid a false track condition, a reacquisition should
be invoked within about 0.5 seconds after DSQM if the initial
subframe is not found.
It is important to suppress faulty branch metrics from a
demodulator so that it does not contaminate the alternate
demodulator. This can happen during a faulty symbol tracking
condition at low SNR. It is not necessarily a problem with a single
(non MRC) demodulator because the signal may be undecodable anyway.
Since MRC combines branch metrics from both demodulators, the
possibility of contamination should be avoided. A DSQM-based
Badtrack detector is described below for this purpose, as well as
for reacquisition.
In addition, filtering should be used as described in the following
sections.
Preacquisition Filtering
To prevent falsely acquiring on large second-adjacent channels,
each primary sideband can be filtered prior to DSQM processing. The
pre-acquisition filter can be designed to provide 60-dB stopband
rejection while limiting the impact on the desired primary
sideband.
An efficient means of computing the DSQM involves decimating the
input complex baseband signal sample rate to approximately 46.5
ksps for each digital sideband (LSB & USB). This can be
accomplished by using the set of isolation filters. However, if the
output sample rate of the digital sidebands is 372 ksps, then a
pair of decimation-by-2 filters can be inserted in front of the
complex mixers and preacquisition filters to provide the expected
sample rate of 186 ksps.
Halfband Highpass Filter
FIG. 11 is a functional block diagram of preacquisition filters
preceded with decimation filters. FIG. 11 shows the complex mixers
320, 322 and preacquisition filters 324, 326 proceeded by a
Halfband Highpass filter 328, 330 to reduce the input sample rate
from 372 to 186 ksps. Complex USB and LSB baseband digital samples
are output from the USB and LSB isolation filters at 372 ksps. A
Halfband Highpass filter is used to decimate the USB or LSB sample
rates from 372 ksps to 186 ksps. The spectrum of this filter has
halfband symmetry, with alternating coefficients equal to zero.
Integer versions of these filter coefficients are presented in
Table 2, showing only one-sided coefficients starting at center
coefficient index 0 through 15. These integer coefficients would be
multiplied by 2.sup.-15 for a unity passband gain. The
negative-indexed coefficients (not shown in Table 2) are equal to
the positive-indexed coefficients.
After decimation-by-2 to 186 ksps, and complex mixing, the USB and
LSB digital sidebands undergo further filtering by the
pre-acquisition filter. This filter should have linear phase and a
minimum output sample rate consistent with passband
characteristics. The upper and lower sidebands should each have a
passband of about 46 kHz, in order to minimize corruption from
first-adjacent analog and second-adjacent digital interference.
This filter can be designed using a decimate-by-4 output sample
rate (46.51171875 ksps).
TABLE-US-00002 TABLE 2 Positive-Indexed Coefficients of Halfband
Highpass USB or LSB Filter. Coefficients 0 through 15 of Halfband
Filter, Starting with Center Coefficient 0 16384 4 0 8 0 2 0 1
-10292 5 -1479 9 -343 3 -34 2 0 6 0 10 0 4 0 3 3080 7 741 11 131 5
4
FIG. 12 shows the spectrum of Highpass Halfband Filter before
decimation-by-2. The output after decimation-by-2 will center the
filter passband to zero Hz. The plots show the undecimated
responses over the Nyquist bandwidth for complex input sample rate
372 ksps, although only the decimated output is computed (for
efficiency). Notice that the baseband 6-dB passband spans the
halfband bandwidth from 93 kHz to 279 kHz, the Nyquist bandwidth at
the output sample rate. The LSB decimation filter has an identical
spectrum, but with negative frequencies. In FIG. 12, the units for
the vertical axis are dB and for the horizontal axis Hz
(frequency); k is a sample index, and K is the total number of
samples in FFT.
Quarterband Pre-Acquisition Filter
The quarterband pre-acquisition filter efficiently isolates a
portion of the output passband of the upper or lower primary
digital sideband filter, suppressing the effects of
adjacent-channel interference. In one embodiment, prior to
filtering, the isolated USB is effectively frequency-shifted by
-155.0390625 kHz, and the isolated LSB is effectively
frequency-shifted by +155.0390625 kHz. The frequency shifting
centers the pre-acquisition filter at baseband (dc), reducing
complexity by allowing a symmetric (real) quarterband filter. In
practice, the frequency shifting can be accomplished by mixing the
baseband alias of the input USB by 31.0078125 kHz (e.sup.j.pi.n/3).
In a similar manner, the baseband alias of the input LSB can be
shifted by -31.0078125 kHz (e.sup.-j.pi.n/3). This frequency
shifting allows the complex phasor to be stored in a circular
lookup table with only 6 coefficients per cycle.
In one example, vectors for complex frequency shifting and filter
coefficients are computed and pre-stored. Pre-store the complex
exponential in a 6-element vector fshft.
.times..pi..times..pi..times..pi..times..pi. ##EQU00014##
The pre-acquisition filter output is further decimated by 4, and is
subsequently used for acquisition. The filter spectrum has
quarterband symmetry, in which every fourth coefficient is zero.
Integer versions of these filter coefficients are presented in
Table 3, showing only positive-indexed coefficients, starting at
center index 0 through 11. These integer coefficients would be
multiplied by 2.sup.-15 for unity passband gain. The
negative-indexed coefficients are equal to the positive-indexed
coefficients.
TABLE-US-00003 TABLE 3 Positive-Indexed Coefficients of Quarterband
Pre-acquisition Filter Coefficients 0 through 11 of Quarterband
Pre-acquisition Filter, Starting with Center Coefficient 0 8192 1
7242 2 4846 3 2080 4 0 5 -912 6 -852 7 -386 8 0 9 130 10 100 11
40
The magnitude spectrum of one embodiment of the pre-acquisition
filter for the USB is shown in FIG. 13. The plots show the
responses over a selected bandwidth within the 372 ksps sample rate
so the filter effects beyond 200 kHz can be seen on the plot. The
actual output of the preacquisition filter is aliased to center the
filter at zero Hz at a sample rate of 46.5 ksps. These plots
include the output spectrum of the upper primary digital sideband
decimation filter, and the effective spectrum of the quarterband
pre-acquisition filter. Notice that the baseband passband spans the
quarterband bandwidth from about 132 to 178 kHz. This passband was
chosen to minimize corruption during acquisition due to
first-adjacent analog FM interference and second-adjacent digital
sideband interference. The LSB characteristics are the same, but
with negative baseband frequencies.
III. DSQM Algorithm Description
The DSQM computation exploits cyclic prefix correlation within each
symbol to construct correlation peaks. The position of the peaks
indicates the location of the true symbol boundary within the input
samples, while the phase of the peaks is used to derive the
frequency offset error over a subcarrier spacing. Frequency
diversity is achieved by independently processing the upper and
lower primary sidebands. When both sidebands are viable, then they
are combined to improve the estimate. An efficient means for
computing the DSQM involves decimating the frequency-shifted input
complex baseband signal sample rate to approximately 46.5 ksps. A
functional block diagram of an embodiment of the quality metric
computation for each sideband is shown in FIG. 14.
FIG. 14 illustrates the USB or LSB quality metric computations. The
inputs to DSQM Processing are symbol-size blocks of upper and lower
primary sideband samples. Each block is comprised of 135 complex
samples at a rate of approximately 46.5 ksps, representing one
symbol time. These blocks have arbitrary boundaries that do not
necessarily coincide with the boundaries of the transmitted
symbols. However, by exploiting a correlation inherent within the
transmitted symbols, their true boundaries can be ascertained.
The input 340 is a 135-sample symbol received from either the upper
or lower sideband preacquisition filter. The input samples are
shifted by 128 samples 342 and the complex conjugate 344 of the
shifted samples is multiplied 346 by the input samples. Sixteen
symbols are folded as shown by block 348 and adder 350. The folded
sums are filtered by a matched filter 352.
The magnitude squared 354 of each input symbol is delayed 342 by
128 samples and added 356 to the current magnitude-squared samples
358. Sixteen symbols are folded as shown by block 360 and adder
362. The folded sums are matched filtered 364. The ratio of the
square of the absolute value of the output of matched filter 352
and the square of the output of the matched filter 364 is computed
as show in block 366 to produce signal Q.sub.m. The index of the
peak value of Q is found as shown in block 368. The complex peak
value is picked and normalized as shown in block 370, and the
result is used for frequency offset estimation.
In the example illustrated by FIG. 14, due to a cyclic prefix
applied at the transmitter, the first and last 6 samples (at 46.5
ksps) of each transmitted symbol are highly correlated. It is
assumed that a zero-value sample is synchronized to the symbol
boundary, so processing of the seventh sample is avoided. DSQM
processing reveals this correlation by complex-conjugate
multiplying each sample in its arbitrary symbol framing with its
predecessor 128 samples away. When the products lie within the
cyclic prefix region of the same transmitted symbol, they form a
6-sample peak with a common phase and amplitude that reflects half
of the complementary root-raised-cosine pulse shape on each end of
the symbol. The location of this correlation peak within the
135-sample product indicates the transmitted symbol boundary, and
the phase indicates the frequency error.
The 6-sample correlation peak over a single symbol is not easily
distinguished from the noisy products of the uncorrelated samples.
To enhance detectability of the peak, the corresponding correlation
products of 16 contiguous symbols are "folded" on top of one
another (pointwise added) to form a 135-sample acquisition vector.
This "conjugate-fold" operation, after initializing vector u to
zeroes, is described as
u.sub.mod(n,135)=u.sub.mod(n,135)+y.sub.ny*.sub.n-128;for n=0,1, .
. . ,135S-1, or equivalently,
.times..times..times..times. ##EQU00015##
where y is the input signal from the preacquisition filter, u is
the folded acquisition vector, m is the folded vector sample index,
s is the folded symbol index, and S=16 is the acquisition block
size (or total number of folded symbols).
The 6-sample folded peak, although visible within the acquisition
vector, is still somewhat noisy. Therefore, the peak is enhanced
with a 6-tap FIR filter h.sub.k whose impulse response is matched
to the shape of the correlation peak.
.times..function. ##EQU00016## .times..times..times. ##EQU00016.2##
where m is the output sample index, u is the matched-filter input,
w is the matched filter output, and h the filter impulse response
defined below.
.function..pi. ##EQU00017## .times..times..times. ##EQU00017.2##
.times..times. ##EQU00017.3##
Notice that this filter is even symmetric with 6 taps, having an
effective group delay of 2.5 samples. This group delay must be
accommodated when locating the symbol-synchronized samples at the
higher non-decimated sample rate.
The correlation peak is enhanced by normalization. Not only is
there a phase correlation between the first and last 6 samples of
an OFDM symbol, but there is also an amplitude correlation due to
the root-raised cosine pulse shaping applied at the transmitter.
This amplitude correlation can be exploited as follows. The
magnitude squared of each input symbol is delayed by 128 samples
and added to the current magnitude-squared samples, as shown in
FIG. 14. After folding the first 16 symbols and matched filtering,
a symbol boundary is apparent. The location of the symbol boundary
is marked by a reduction in amplitude of the resultant waveform.
Normalization of the existing correlation peak with this waveform
enhances the peak by reducing the level of all samples except those
coincident with the symbol boundary. This operation, after
initializing vector v to zeroes, is described as
v.sub.mod(n,135)=v.sub.mod(n,135)+|y.sub.n|.sup.2+|y.sub.n-128|.sup.2;for
n=0,1, . . . ,135S-1, or equivalently,
.times. ##EQU00018## .times..times..times. ##EQU00018.2##
where y is the input signal from the preacquisition filter, v is
the folded vector, m is the folded vector sample index, s is the
folded symbol index, and S=16 is the acquisition block size (or
total number of folded symbols).
The 6-sample folded peak, although visible within the acquisition
vector, is still somewhat noisy. Therefore, the vector v is
enhanced with a 6-tap FIR filter g.sub.k whose impulse response is
matched to the shape of the symbol boundary region.
.times..function. ##EQU00019## .times..times..times.
##EQU00019.2##
The matched filtering of the normalization waveform is identical to
that performed for the correlation peak, except the matched filter
taps are squared and then halved to ensure proper
normalization:
##EQU00020## .times..times..times. ##EQU00020.2## .times..times.
##EQU00020.3## where k is the index of taps in the matched filters,
h.sub.k are the existing taps for the conjugate-multiplied
correlation peak, and g.sub.k are the new taps for the
normalization waveform. Notice that this filter is also even
symmetric with 6 taps, having an effective group delay of 2.5
samples. This group delay must be accommodated when locating the
symbol-synchronized samples at the higher non-decimated sample
rate. A quality metric vector Q is computed from vectors w and
x.
##EQU00021## .times..times..times. ##EQU00021.2##
The peak value Q.sub.P of the vector Q, and its index P, are
identified. The peak value Q.sub.P is further processed to reduce
the probability of false detection due to a spur, for example. A
strong spur in the absence of noise or digital signal could produce
a correlation peak that is greater than one over the entire symbol
correlation vector. To prevent this false detection, conditions are
placed on the Q.sub.P result, which would zero the Q.sub.P if a
false detection is suspected. One condition is that the peak
Q.sub.P value must be less than one. A second condition is that the
sum of the correlation samples, spaced every 3 samples away from
the peak sample, must be less than some value (for example, this
sum must be less than 2). This discrimination is implemented by
multiplication of the peak value Q.sub.P by two Boolean (0 or 1
value) expressions.
<.times..times.< ##EQU00022##
The peak value of the normalized correlation waveform is
representative of the relative quality of that sideband. The entire
computation just described in this section is done for both the USB
and the LSB, and the final results are saved as Q.sub.U, Q.sub.L,
P.sub.U and P.sub.L for subsequent DSQM computation.
Once the correlation waveform is effectively normalized for each
sideband, the value and index of the peak are found. The peak index
delta compares the peak indices of the upper and lower sidebands
for each sixteen-symbol block. Since the symbol boundaries are
modulo-135 values, the computed deltas must be appropriately
wrapped to ensure that the minimum difference is used.
.DELTA.P=min{|P.sub.U-P.sub.L|,135-|P.sub.U-P.sub.L|} where P.sub.U
and P.sub.L are the peak indices of the normalized correlation
waveform for the upper and lower sidebands.
DSQM Calculation
Once the peak index delta and quality estimates have been computed,
they are used to calculate the DSQM. The DSQM separately examines
the quality of each individual sideband, in addition to evaluating
the peak index delta and sum of the quality estimates from both
sidebands. In this way, a viable signal can be successfully
identified even when one of its sidebands has been corrupted by
interference.
False detections may occur on an analog-only signal in a very
low-noise channel. In this case, some of the FM signal components
exist in the DSQM detection band, and can trigger DSQM detection.
The correlation on upper and lower sidebands is unlikely to peak at
the same location, and this false detection would more likely occur
on one sideband only.
A temporal consistency check can be used to discriminate against
this condition. This temporal consistency check prevents initial
detection on one sideband only. If only one sideband passes the
threshold on the first DSQM measurement, and .DELTA.P>1, then a
second DSQM computation is used to assess if the correlation peak
occurs at the same location (P.sub.L or P.sub.U) on that sideband.
If the peak index from a sideband is consistent on two consecutive
DSQM measurements, then the acquisition is declared successful.
The flowchart of FIG. 15 can be used for acquisition of the digital
signal. It can also be used in a seek function, coordinated by the
host controller, for example. The seek signal quality threshold
SeekThres is usually set to a higher level than that used for
acquisition, so that the seek function stops on only a reasonably
good signal. The normal acquisition threshold Thres is lower to
allow acquisition on marginal signals. Seek or acquisition is
determined to be successful or not after one or two iterations. The
algorithm continues to iterate until the digital signal is
successfully acquired, or is interrupted by the host controller,
for example.
The first iteration of the ACQ algorithm 380 is indicated by
initializing the InitFlag to one, and DSQMSeqNum to one (block
382). Quality metrics Q.sub.L and Q.sub.U, peak index indices
P.sub.L and P.sub.U, and .DELTA.P are computed (block 384). If the
initial flag is not equal to 1 (block 386), a Temporal Consistency
Check is performed on each sideband (block 388), except on the
first iteration when previous Peak indices are not available. If
the initial flag is equal to 1, a DSQM is computed and selected
(maximum Q, if .DELTA.P<2); upper and/or lower sideband(s) are
identified by setting L.sub.acq=1 and/or U.sub.acq=1 (block 390).
If this DSQM value exceeds the acquisition threshold Thres (e.g.,
Thres=0.2) (block 392), then the DSQM value, along with DSQMSeqNum
and DSQMDetBit, are output (block 394). The DSQMDetBit is
determined by Boolean result of comparing the DSQM to the Seek
Threshold (e.g., 0.5). DSQMDetBit=DSQM>SeekThres
If the first DSQM (of this iteration) fails to exceed the
acquisition threshold (e.g., 0.2), then another DSQM is computed
based on the maximum of Q.sub.L or Q.sub.U (but not both together)
(block 396). If this DSQM fails to exceed the acquisition threshold
(block 398), then this DSQM value, DSQMSeqNum and DSQMDetBit are
output (block 400) and DSQMSeqNum is incremented (block 402); the
next iteration of the algorithm is then executed using the
parameters in block 403. However, if this DSQM exceeds the
acquisition threshold, then successful acquisition is declared
(block 404) (if this is not the first algorithm iteration, block
406); otherwise, the next algorithm iteration can resume.
IV. Frequency & Timing Acquisition Example
Acquisition is the process of establishing initial symbol
synchronization and frequency offset for subsequent tracking. A
threshold for DSQM is established where a sufficiently reliable
signal is detected. The symbol timing sample is determined by the
peak quality index P. This index is determined with a decision rule
based on which sidebands were used to yield the final DSQM value.
The selected sidebands are indicated by the Boolean values of Lacq
and Uacq, as determined in the algorithm of FIG. 15. If either the
USB or LSB alone were used, then the peak index would be the index
of the selected sideband. However, if both sidebands were used,
then the indices are averaged modulo 135. Adjustments to this value
for decimation, filter delays, or other implementation delays must
be performed.
.times..times..times..times..times..times.<.times..times.>
##EQU00023##
The frequency offset in Hz can be estimated using the complex value
of the normalized correlation peak. The value for each sideband can
be phase-adjusted to accommodate the frequency-shifting from the
center of the preacquisition bandwidth. The final value depends on
which sideband(s) is used, per the following expression:
e.pi. ##EQU00024## e.pi. ##EQU00024.2##
.times..times..times..times..times..times. ##EQU00024.3##
The frequency error in Hz is proportional to the angle of
Qcmplx.
.times..times..times..times..pi..times..function..function.
##EQU00025##
However the NCO may require the negative of this frequency error to
be translated into a phase increment phinc in radians per
sample.
.pi. ##EQU00026## where f.sub.s is the NCO sample rate.
Furthermore, it is common for fixed-point implementations to use
the modulus range of a two's complement number to represent a full
circle.
V. DSQM-Lite for Antenna Diversity
While a digital signal quality metric (DSQM) can be used for
antenna diversity switching, the DSQM computed during signal
acquisition is computationally intensive, and involves redundant
processing after symbol synchronization is established. This
section describes an algorithm for more efficient DSQM computation,
called DSQM-Lite. It is derived from the acquisition algorithm
described in previous sections, but the computational complexity is
reduced by taking advantage of symbol synchronization. Since the
locations of the cyclic prefix regions of the symbols are known,
there is no need to compute all the correlation points across the
entire symbol.
This DSQM function is based on the previously described DSQM
technique to generate an appropriate metric that can be used for
antenna diversity switching (among other uses). The DSQM algorithms
process groups of 16 symbols to produce a metric.
Since the symbol boundaries are already established in the symbol
dispenser when the DSQM is used for diversity switching, there is
no need to compute more than one peak, and the phase information is
not needed or used here.
Efficient Computation of DSQM for Diversity Switching
After initial acquisition as described in Section IV above, a
substantial reduction in MIPS (millions of instructions per second)
can be realized by limiting the processing of signal samples to the
cyclic prefix regions of the symbols. Since the symbol samples are
already framed by the symbol dispenser in the present
implementation, it is relatively straightforward to select the
cyclic prefix regions for DSQM processing. There are only 6 samples
to process at each end of the 135-sample symbol at the
decimate-by-16 sample rate. It will be shown later that only sample
indices 1 through 6 and 129 through 134 need to be computed; sample
0 is not needed since it should be synchronized to have a zero
value.
DSQM-Lite Computation
A process for computing DSQM-Lite uses the following steps.
STEP 0 (for 372 ksps input): If the input sample rate is not
approximately 186 ksps, then a decimation filter can be inserted to
achieve that sample rate. A Halfband Highpass decimation filter was
previously described for this purpose; however, it is more
efficient to compute only the samples needed for this DSQM-lite
computation. Define USB2x and LSB2x as the 1080-sample input symbol
vectors at 372 ksps. These vectors are filtered by the 31-element
Halfband Highpass filter hbf, while decimating by 2 to yield
540-element vectors USB and LSB. Notice that although 540-sample
vectors are generated, only the range n=1 to 35 and n=505 to 539
need be computed, and the remaining uncomputed elements are set to
zero.
.times..times..times..times. ##EQU00027##
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times. ##EQU00027.2##
.times..times..times..times. ##EQU00027.3##
.times..times..times..times..times..times..times..times..times..times..ti-
mes..times..times..times. ##EQU00027.4##
Uncomputed elements n=0 and n=36 through 504 are not used in
subsequent calculations. Filter coefficients for hbf are scaled by
2.sup.-15 for fixed point implementation.
##STR00001##
STEP 1: Place the frequency-shifted endpoints pshft and qshft for
the upper and lower sidebands in vectors for each symbol:
.function..times..times.>.times..times..function..times..times.<.ti-
mes..times..function..times..times.>.times..times..function..times..tim-
es.<.times..times..times..times..times..times..times..times..times..tim-
es..times..times..times..times..times..times..times..times..times.
##EQU00028##
STEP 2: These vectors are filtered with hqb, and then decimated by
a factor of 4.
.times. ##EQU00029## .times. ##EQU00029.2## .times. ##EQU00029.3##
.times. ##EQU00029.4## .times..times..times. ##EQU00029.5##
STEP 3: The "conjugate multiply and fold" operation is
mathematically described for each upper or lower sideband by the
following equations:
.times. ##EQU00030## .times. ##EQU00030.2## .times..times..times.
##EQU00030.3## where s is the folded symbol index, and S=16 symbols
is the acquisition block size.
STEP 4: The normalization factor v is used to scale the DSQM to a 0
to 1 range:
.times. ##EQU00031## .times. ##EQU00031.2## .times..times..times.
##EQU00031.3##
STEP 5: The quality Q value for either the lower or upper sideband
is computed as:
.times..times. ##EQU00032## .times..times. ##EQU00032.2## where
filter coefficients h and the g are precomputed as:
.times..times. ##EQU00033##
STEP 6: Finally, the composite DSQM metric is computed
(0<DSQM<1). Notice that the additional sample timing
condition is not imposed when the two sidebands are combined. This
is because symbol synchronization is assumed and the timing
alignment is ensured by the symbol dispenser. DSQM=max
[Q.sub.U,Q.sub.L,min{1,Q.sub.U+Q.sub.L-0.2}]
Next, the effect of DSQM quality threshold (Q) and peak sample
delta (AP) threshold is examined. Simulation results were used to
estimate the probability of an individual (e.g., one sideband) DSQM
symbol timing error (135 samples/symbol) as a function of Cd/No.
For the purpose of this analysis it is assumed that the timing
error is relative to zero, and is defined over 135 samples ranging
from -67 to +67. Negative timing errors have the same probability
as positive timing errors, so they are not shown in the tables
below. The timing error is assumed uniform outside of .+-.5
samples, the cyclic prefix region.
Tables 4 through 7 show the conditional probability Psample(P,
thres, Cd/No) that P is a particular one of the 135 possible
values, given that the quality threshold (i.e., 0.0, 0.1, 0.15, and
0.2) is exceeded (Q>thres) as characterized through simulation.
The variable Cd/No is the carrier to noise density ratio in units
of dB_Hz. Although a greater thres value discriminates against
erroneous peaks, it also reduces successful acquisition probability
for each DSQM trial. Notice that Table 4 imposes no quality
condition on DSQM quality threshold since thres=0.0 in this
case.
TABLE-US-00004 TABLE 4 Probability of timing error when DSQM >
0.0, Psample(P, 0.0, Cd/No) Timing Cd/No = Cd/No = Cd/No = Cd/No =
Cd/No = Error P 50 dB- 51 dB- 52 dB- 53 dB- 54 dB- (samples) Hz Hz
Hz Hz Hz 0 0.175 0.285 0.410 0.524 0.635 1 0.102 0.134 0.163 0.168
0.162 2 0.029 0.028 0.024 0.014 0.006 3 0.010 0.0093 0.0051 0.0035
0.001 4 0.0059 0.0038 0.0028 0.00088 0.0005 5+ 0.0042 0.0029 0.0016
0.00081 0.00020
TABLE-US-00005 TABLE 5 Probability of timing error when DSQM >
0.1, Psample(P, 0.1, Cd/No) Timing Cd/No = Cd/No = Cd/No = Cd/No =
Cd/No = Error P 50 dB- 51 dB- 52 dB- 53 dB- 54 dB- (samples) Hz Hz
Hz Hz Hz 0 0.256 0.383 0.480 0.562 0.649 1 0.151 0.171 0.183 0.178
0.163 2 0.038 0.033 0.023 0.014 0.0053 3 0.012 0.0092 0.0048 0.0028
0.0011 4 0.0046 0.0030 0.0016 0.00088 0.00026 5+ 0.0026 0.0015
0.00075 0.00038 0.0000900
TABLE-US-00006 TABLE 6 Probability of timing error when DSQM >
0.15, Psample(P, 0.15, Cd/No) Timing Cd/No = Cd/No = Cd/No = Cd/No
= Cd/No = Error P 50 dB- 51 dB- 52 dB- 53 dB- 54 dB- (samples) Hz
Hz Hz Hz Hz 0 0.427 0.503 0.538 0.605 0.672 1 0.189 0.205 0.201
0.184 0.158 2 0.028 0.022 0.019 0.0094 0.0039 3 0.012 0.0043
0.00040 0.0012 0.00033 4 0.0014 0.0014 0 0 0 5+ 0.00088 0.00025
0.00017 0.000052 0.000018
TABLE-US-00007 TABLE 7 Probability of timing error when DSQM >
0.2, Psample(P, 0.2, Cd/No) Timing Cd/No = Cd/No = Cd/No = Cd/No =
Cd/No = Error P 50 dB- 51 dB- 52 dB- 53 dB- 54 dB- (samples) Hz Hz
Hz Hz Hz 0 0.477 0.565 0.550 0.609 0.701 1 0.215 0.199 0.203 0.188
0.147 2 0.031 0.016 0.021 0.0054 0.0027 3 0.0077 0 0 0.0011 0 4 0 0
0 0 0 5+ 0.00012 0.000049 0.000019 0.0000085 0
The probability of a particular timing error when no signal is
present is independent of thres, and is simply the uniform
probability of selecting any one of the 135 sample timing
offsets.
.function..times..apprxeq. ##EQU00034##
The probability that one DSQM quality measurement exceeds the
threshold (Q>thres) for a given Cd/No is defined as P1(thres,
Cd/No). This parameter is a function of thres and Cd/No as
characterized through simulation; results are shown in Table 8.
TABLE-US-00008 TABLE 8 Probability of exceeding DSQM threshold,
P1(thres, Cd/No) Cd/No = Cd/No = Cd/No = Cd/No = Cd/No = DSQM No 50
51 52 53 54 thres signal dB-Hz dB-Hz dB-Hz dB-Hz dB-Hz 0.10 0.261
0.434 0.542 0.724 0.851 0.947 0.15 0.022 0.093 0.175 0.316 0.530
0.760 0.20 0.0015 0.016 0.040 0.103 0.232 0.470
Acquisition Using Multiple DSQM Measurements
The acquisition probability is the joint probability that a pair of
peak indices (e.g., P.sub.U or P.sub.L) is within D samples of each
other, given that at least one quality measurement exceeds thres.
This can be analyzed from the sample timing offset Psample data.
For given values of thres and Cd/No, this probability is computed
using the probabilities in Tables 3 through 7.
.function..times..times..times..times..times..times..times..function..tim-
es..function..times..times..times..times..function..times..function.
##EQU00035##
The expression above indexes the sample timing data for Psample
slightly above 67 samples and below -67 samples. In these cases
there is a modulo wrap-around where sample 68 is equivalent to -67,
and sample -68 is equivalent to 67. However, since the probability
is uniform outside of .+-.4 samples, these values of Psample are
held constant. So any index for Psample outside of .+-.4 has an
index equivalent to P=5. Notice that Pacq does not guarantee that
all acquisitions have an acceptable sample timing error, although
the pair of timing P measurements is within D samples of each
other. So the probability Pacq includes a small fraction of
acquisitions where the sample timing error is faulty, leading to a
"Badtrack" condition.
If the initial symbol timing error is greater than 3 samples, then
the symbol tracking will likely converge to a faulty stable sample
offset (about 27 samples error), resulting in a "Badtrack." The
Badtrack condition sometimes occurs at low SNR when the error is 3
samples, and the symbol clock frequency error could also affect the
probability of Badtrack. Therefore a more conservative condition
requiring less than 3 samples timing error for proper tracking is
analyzed. The probability of Badtrack is conditioned on passing the
requirements for detecting acquisition, but the timing estimate
error is outside of .+-.2 samples. The probability of Badtrack is
expressed as
.function..times..times..times..times..times..times..times..function..tim-
es..function..times..times..times..times..function..times..function.
##EQU00036##
Then the probability of a good acquisition where the timing falls
within -2.ltoreq.P.ltoreq.2 is:
Pgoodacq(D,thres,Cd/No)=Pacq(D,thres,Cd/No)-Pbadtrack(D,thres,Cd/No).
Since Pbadtrack is generally much smaller than Pacq for the cases
examined here, Pgoodacq is only slightly smaller than Pacq.
FIG. 16 shows the probability of a good acquisition where the DSQM
timing error is P.ltoreq.2 samples. FIG. 17 shows the probability
of a bad acquisition where the DSQM timing error is P>2
samples.
The probability of false alarm occurs when no signal is present, at
least one of the pair of DSQM quality measurements exceeds the
threshold, and .DELTA.P.ltoreq.D. This is equivalent to the
expression for Pacq when no signal is present.
.function.>.times..function. ##EQU00037##
which can also be expressed as:
.function..times..times..times. ##EQU00038##
TABLE-US-00009 TABLE 9 Probability of false alarm (.DELTA.P
.ltoreq. D) with no signal, Pfalsealarm (D, thres) thres D = 1 D =
2 0.00 0.022 0.037 0.10 0.010 0.017 0.15 0.001 0.0016 0.20 0.000067
0.00011
Temporal Consistency
The temporal consistency check is intended for acquisition of a
signal where one sideband is severely corrupted. The corrupted
sideband is assumed to yield an unreliable symbol timing offset
value P, so the .DELTA.P.ltoreq.D condition is unlikely to be
satisfied. The DSQM decision rule requires a timing consistency
within .DELTA.P.ltoreq.D samples for a pair of DSQM measurements.
This pair of DSQM measurements normally consists of the upper and
lower sideband values. However, if this consistency is not met for
the pair of sidebands, then the next pair of DSQM samples is also
used to check for temporal consistency on the same sideband.
Furthermore, this temporal consistency requires only the most
recent DSQM value to exceed the threshold, while the previous value
on the same sideband must be within .+-.D samples. There is no
requirement that the previous value exceeds the threshold. This
temporal consistency condition will tend to increase all the
acquisition probabilities (Pacq, Pbadtrack, Pgoodtrack and
Pfalsealarm), especially for low values of Cd/No. For low Cd/No in
AWGN, these probabilities are expected to nearly double due to the
temporal consistency check. The doubling can be explained by
allowing an extra check for .DELTA.P.ltoreq.D on the same sideband
in addition to the alternate sideband. For low Cd/No, only one
sideband is likely to exceed the threshold. Then the probability of
a false alarm, including the temporal consistency check, is
modified to approximately
.function..times..times..times. ##EQU00039## Table 10 is similar to
Table 9, except the probability of false alarm includes the
temporal consistency check.
TABLE-US-00010 TABLE 10 Probability of false alarm (.DELTA.P
.ltoreq. D) with temporal consistency, no signal, Pfalsealarm (D,
thres) thres D = 1 D = 2 0.00 0.044 0.074 0.10 0.020 0.034 0.15
0.002 0.0032 0.20 0.000134 0.00022
Frame Synchronization Analysis
Frame synchronization is described here as a two-step process:
Initial Subframe Found, followed by Subframe Lock. This process
starts after a successful DSQM detection identifies the symbol
timing offset, and the symbol-synchronized OFDM demodulation
commences. However, if an Initial Subframe is not detected after a
predetermined time following DSQM acquisition, then a reacquisition
must be initiated to prevent sample timing drift caused by clock
frequency error.
To detect the Initial Subframe, the receiver performs a sliding
correlation over all the OFDM subcarriers and the received OFDM
symbols. The correlation is for an 11-bit sync pattern spread over
a 32-symbol Subframe in all of the Reference Subcarriers. A
subcarrier correlation is declared when all 11 sync bits match the
sync pattern for that subcarrier. Initial Subframe Found is
declared when correlation is successful on a predetermined number
of subcarriers spaced a multiple of 19 subcarriers apart, and on
the same 32-bit subframe. When Initial Subframe Found occurs, the
32-bit subframe boundaries are established, as well as the location
of the Reference Subcarriers. If the Subframe is not found after a
predetermined time after DSQM acquisition, then this process is
terminated, and a reacquisition is initiated.
Subframe Lock is established when another predetermined number of
subcarrier correlations occurs on the established Reference
Subcarriers, and are spaced from the Initial Subframe Found by an
integer multiple of 32 symbols. If the symbol tracking is initiated
immediately after Initial Subframe Found, then it may not be
necessary to place a time limit on Subframe Lock before a
reacquisition. This is because further symbol timing drift is
prevented by symbol tracking.
The following analysis characterizes the probabilities associated
with Initial Subframe Found and Subframe Lock. This can be combined
with the DSQM analysis to determine the probabilities of successful
acquisition, faulty acquisition, and estimates of the time required
for Subframe Lock.
First compute correlation probability on one subcarrier with 11
sync bits. Because of the possibility of large phase errors due to
initial symbol error (before symbol tracking converges), one must
consider the 4 possible phases of the signal (I,Q and complements)
for correlation possibilities. Note that this 4-phase detection
method may introduce other error conditions when 2 phases straddle
the boundary. This probability is approximated by:
Psync(BER)=1-[1-(1-BER).sup.11](1-BER.sup.11)(1-0.5.sup.11) where
the probability of bit error (BER) for differentially detected
BPSK, or DBPSK is
e ##EQU00040##
For the BPSK reference subcarrier of the IBOC signal, the
relationship between Eb and Cd in dB is:
Eb.sub.dB=Cd.sub.dB-51dB.
The quantity Cd/No is expressed in units of dB_Hz. Then the BER can
be expressed as a function of Cd/No.
e.function. ##EQU00041##
In order to compute the probability of Initial Subframe Found and
Subframe Lock, some intermediate probabilities are computed. The
probability that a successful correlation occurs on Nsc
subcarriers, given that the Primary reference subcarriers are
already identified, and are synchronized to the Subframe boundary
is:
.function..times..function..function. ##EQU00042##
The probability Psf is also the conditional probability of Subframe
Lock in any one Subframe time (32 symbol period), given that
Initial Subframe Found is successful.
Allowing for all 19 possible Reference Subcarrier offsets in a
partition, and for all 32 symbol possibilities in a Subframe, the
average probability of Initial Subframe Found over every 32-symbol
shift of a subframe is:
Pfound(Nsc,BER)=1-[1-Psf(Nsc,BER)][1-Psf(Nsc,0.5)].sup.1932-1.
The average time (seconds) required for Initial Subframe found,
given that the signal is acquired and no reacquisitions are
allowed, can be computed as:
.function..function. ##EQU00043## where fsym is the OFDM symbol
rate. The plot of FIG. 19 shows Tfound(Nsc,BER) for Reference
Subcarrier correlation thresholds of 4, 3, and 2 over a range of
Cd/No.
FIG. 18 is a plot showing the average time required for Subframe
lock after Initial Subframe found. This assumes no
reacquisitions.
The average time (seconds) required for Subframe Lock, given
Initial Subframe Found, can be computed as:
.function..function. ##EQU00044##
The plot of FIG. 19 shows Tsf(Nsc,BER) for Reference Subcarrier
correlation thresholds of 4, 3, and 2 over a range of Cd/No.
The probability of Initial Subframe found over the allotted Nsf1
Subframe periods to find the sync pattern is:
PfoundNsf(Nsc1,BER,Nsf1)=1-(1-Pfound(Nsc1,BER)).sup.Nsf1.
The probability of Subframe Lock over the allotted Nsf2 Subframe
periods is: PlockNsf(Nsc2,BER,Nsf
2)=1-(1-Psf(Nsc2,BER)).sup.Nsf2.
Selection of Parameter Values for Frame Sync
Based on the probability analyses in the previous sections, the
following parameter values are recommended:
D=1 Sample offset difference (.DELTA.P.ltoreq.D) permitted
thres=0.1 DSQM quality threshold for acquisition
Nsc1=3 Number of sync correlations required for Initial Subframe
Found
Nsc2=2 Number of sync correlations required for Subframe Lock
Nsf1=4 Number of Subframes for Initial Subframe Found before
reacq
Nsf2=4 Number of Subframes for Subframe Lock before reacq
False Acquisition And Subframe Lock Rate
It was shown in the DSQM analysis that the probability of false
DSQM acquisition (with no signal) Pfalsealarm is approximately 0.02
(thres=0.1, D=1, including temporal consistency check) for every
16-symbol period. Then the average time between false DSQM
acquisitions is:
.function..times..times. ##EQU00045##
The probability of Initial Subframe Found (no signal) within the 4
Subframes allotted is: PfoundNsf(3,0.5,4)=0.027.
The time period allotted (Nsf Subframes) for Initial Subframe Found
is:
.function..times..times..function..times..times. ##EQU00046## and
TNsf(4)=0.372seconds.
The average time required for faulty Initial Subframe Found, given
a faulty DSQM acquisition (BER=0.5) is:
.function..times..times..times..times..function..times..times..function..-
times..times..times..times. ##EQU00047## .function..times..times.
##EQU00047.2##
The probability of faulty Subframe Lock over the allotted Nsf=4
Subframes, given a faulty Initial Subframe found, is:
PlockNsf(2,0.5,4)=3.410.sup.-3.
The time period allotted (Nsf Subframes) for Subframe Lock in this
case is the same as for Initial Subframe Found: TNsf(4)=0.372
seconds.
The average time required for faulty Subframe Lock, given a faulty
DSQM acquisition (BER=0.5) and faulty Initial Subframe Found
is:
.function..times..times..times..times..function..times..times..times..tim-
es..function..times..times..times..times. ##EQU00048##
.function..times..times..times..times..times..times..times..times.
##EQU00048.2##
Then false Subframe Lock occurs about once in 8 hours with no
signal present. However, it is also assumed that symbol tracking
doesn't result in a false lock, which is influenced by the sample
clock error (e.g., up to 100 ppm). A combination of large clock
error and initial sample offset error from DSQM could result in a
false lock in symbol tracking, or "Badtrack". A means of detecting
Badtrack should be implemented.
VII. DSQM-Lite for Badtrack Detection
A Badtrack detection method is needed to prevent the demodulator
from remaining in a stuck condition while outputting faulty branch
metrics. Badtrack is the result of the symbol tracking being stuck
at a faulty sample offset (e.g. 27 samples error). This is due to a
2-.pi. phase shift (instead of 0) between adjacent Reference
Subcarriers. The Badtrack is especially important in an MRC
diversity receiver where each demodulator can operate at a lower
SNR, and contamination of one demodulator in a Badtrack state to
the other demodulator is possible. A reacquisition is invoked when
a Badtrack is detected. The existing fourth-power Badtrack
detection method is unreliable for Cd/No<54 dB_Hz. However a
DSQM_lite-based detection method is more reliable, and is described
here. The DSQM_lite function provides periodic digital signal
quality metrics (every 16 or 32 symbols), but requires fewer MIPS
than the original DSQM function. Fewer MIPS are needed because it
exploits knowledge of the location of the cyclic prefix region
after initial acquisition.
Assume DSQM_lite samples are available every 16-symbol period.
These can be filtered with a unity-gain lossy integrator with a
time constant of about 8 samples. At the start of DSQM_lite
filtering, the filter memory DSQM_lite jilt should be initialized
to DSQM_lite_filt_init (e.g., 0.08), which is between the two
threshold values for Badtrack detection and low signal suppression
described later in this section. The filter initialization (instead
of zero) reduces the initial period when a good signal is
suppressed due to filter time constant. The DSQM_lite IIR filter is
a unity-gain lossy integrator with a time constant of about 8
DSQM_lite samples (128 symbols). The filter expression is:
DSQM_lite_filt.sub.n=0.875DSQM_lite_filt.sub.n-1+0.125DSQM_lite.sub.n.
Branch metrics can be suppressed (zeroed) when
DSQM_lite_filt<thres_nosig. (e.g., thres_nosig=0.1) The
DSQM_lite_filt value approaches approximately 0.15 for Cd/No=51
dB_Hz, the minimum expected operating value. if
DSQM_lite_filt.sub.n<thres_nosig;then ZERO all Branch
metrics.
A counter is incremented when the filtered
DSQM_lite_filt<thres_badtrack (e.g., thres_badtrack=0.06). This
threshold value offers sufficient margin for Badtrack detection
since DSQM_lite_filt approaches approximately 0.03 in a Badtrack
condition or when no signal is present. This should be effective in
preventing contamination to the alternate demodulator in the MRC
case.
Reacquisition is invoked when the counter indicates a sufficiently
long duration. The counter is initialized to zero at DSQM
acquisition, and reset to zero whenever the filtered
DSQM_lite_filt.gtoreq.thres_badtrack.
.times..times..times.<.times..times..times..times.>.times..times..t-
imes..times..times..times..times..times..times..times.
##EQU00049##
FIG. 20 shows a plot of DSQM_lite_filt versus time (in DSQM
periods) for Cd/No=51 dB_Hz. The horizontal axis units are in DSQM
samples, where each sample spans 16 symbols (46.5 msec). The
average value approaches about 0.15 in this case. FIG. 21 shows a
plot, of DSQM_lite_filt versus time (in DSQM periods) for no signal
present (noise only). The average value approaches about 0.03 when
no signal is present. The horizontal axis units are in DSQM
samples, where each sample spans 16 symbols (46.5 msec).
FIGS. 22 through 24 show DSQM_lite_filt at 51 dB_Hz with different
values of symbol timing error. The symbol tracking was disabled in
these cases, and the symbol timing error was held constant. The
degradation due to symbol timing error can be assessed by comparing
the DSQM_lite_filt value to FIG. 20. FIG. 20 shows that the
DSQM_lite_filt approaches approximately 0.15 when no sample error
is present. FIGS. 22 through 24 show that the DSQM_lite_filt
approaches approximately 0.12, 0.08 and 0.05 with sample offset
errors of 4, 8, and 12 samples, respectively, at 540
samples/symbol. The BER (after FEC decoding) measured at 8 samples
offset is approximately 0.5 for a single (non-MRC) modem,
indicating that the branch metrics may provide insignificant
improvement for MRC combining. That is why the DSQM_lite_filt
threshold for branch metric suppression is set at the particular
value of thres_nosig.
VIII. Implementation Considerations
Since the pair of digital demodulators may not be in the same state
(e.g. reacq, frame sync, valid branch metrics) at the same time, an
arbitration scheme must be developed. One possibility is that both
digital demodulators (D0 and D1) operate mostly autonomously from
each other. The first demod to reach Subframe Lock shall coordinate
operations (master) for combining branch metrics, and downstream
(deinterleaving, decoding, etc.). Branch metrics can be combined
from alternate demods when available. It is assumed that the
demodulation process is multiplexed by alternating symbol
processing for the pair of demodulators. Then only one demodulator
at a time will change the state. Transitions between states can be
initiated either by a reacquisition (reacq) or a Subframe Lock
(SFLock). Each demodulator can be in only one of two modes, SYNC or
DECODE. For each demodulator the SYNC mode is entered by a reacq,
and the DECODE mode is entered by a SFLock.
FIG. 25 is a state diagram for MRC coordination and arbitration.
There are 4 possible states for the MRC Arbitration State diagram
shown in FIG. 25. The state is determined by the pair of
demodulator modes.
The downstream functions (deinterleaving, decoding, etc.) are
initialized in State 0. Upon entering State 0, the deinterleaver is
not receiving symbols from either of the demodulators, since they
are both in SYNC mode. The first demodulator to establish Subframe
Lock initiates the downstream functions. The last demodulator to
enter SYNC mode disables the downstream functions.
The described acquisition and tracking modifications will allow
more reliable acquisition and tracking at lower SNRs to aid MRC
performance. Reducing DSQM threshold from 0.2 to 0.1 will improve
acquisition time at low SNR.
All fourth-power-based processing has been eliminated, including
symbol tracking and bad-track detection/reacquisition control.
The symbol tracking algorithm is disabled until Initial Subframe
Found. The symbol sample offset correction determined by DSQM is
maintained. The sample timing may drift due to the difference in
transmitter and receiver clocks (e.g., 100 ppm will drift 18.6
samples/second at 186 ksps). The symbol tracking is intended to
prevent further sample error drift after Initial Subframe Found.
One sample at a time is corrected. The receiver allows the sample
slip to drift for a limited time until it is out of symbol tracking
range. If it drifts any longer then a reacquisition is
performed.
Before Initial Subframe Found, the symbol tracking loop input and
symbol tracking SNR should be 0. After Initial Subframe Found,
symbol tracking on reference subcarriers can begin.
Filtered sync weights can be used immediately upon starting the
symbol tracking loop. All canned weights (4.sup.th power and pilot)
can be deleted. Initializing sync weights to canned weights instead
of zero can be considered.
The fast-track period, with the symbol tracking loop gain
collapsing from 0.2 to 0.02, can also start immediately after
Initial Subframe Found. It can remain 400 symbols long. However,
other actions previously performed during fast-track are deleted.
Since tracking on pilots, the symbol tracking error input is scaled
by 1/19. The symbol tracking error input is clipped to .+-.1 (it
was previously clipped to .+-.5 during fast-track). The SNR-based
flywheel gain shall be set to 1 during the fast-track period (until
proportional gain=0.02).
Disable all SNR-based reacquisition conditions. Note that SNR=0
until 21 symbols after Init Subframe Found. In Initial Subframe
Detection state, remove reacq if SNR<0.1 and no subframes
detected within 100 symbols. In Subframe Verification state, remove
reacq if 125 symbols have been processed since entering this state,
and SNR<0.1.
The rules for determining Subframe Found and Lock have been
modified. The Initial Subframe Found requires only three 11-bit
sync correlations spaced by 19 subcarriers. If not detected within
128 symbols (4 Subframe periods) after successful DSQM, then
perform a reacquisition. The Subframe Lock checks only the
identified reference subcarriers from initial subframe correlation
at multiples of the 32-symbol spacing.
Only the current subframe spacing needs to be checked against the
initially detected subframe, not all previously detected subframes.
The 32-subframe array can be removed.
The second subframe requires only two 11-bit sync correlations. If
Subframe lock is not established within 128 symbols (4 Subframes)
after Initial Subframe Found, then perform a reacquisition.
The Reference Subcarrier ID (coarse bin offset) can be checked for
consistency between the Initial Subframe Found and 2.sup.nd
detected subframe, before declaring Subframe Lock. A reacquisition
can be performed if the Reference Subcarrier IDs are different.
Bad-track detection can be implemented using IIR-filtered
DSQM_lite_filt, replacing fourth-power bad-track detection. A
reacquisition can be invoked when bad-track is detected.
DSQM can be calculated every 16 symbols. IIR can be a single pole
unity-gain lossy integrator with alpha=1/8. Filter can be
initialized to DSQM_lite_filt_init (e.g., 0.03 to 0.08).
A counter can be incremented when the filtered
DSQM_lite_filt<thres_badtrack (e.g., 0.06). A reacquisition can
be invoked when the counter exceeds 100 DSQM periods (1600
symbols). The counter can be reset to 0 when filtered DSQM-lite
exceeds thres_badtrack. The timeout after Subframe lock may be
increased.
Whenever the filtered DSQM_lite_filt<thres_nosig (e.g.,
0.6.ltoreq.thres_nosig.ltoreq.0.1), all branch metrics can be
zeroed. The only original Synchronization Control Reacq condition
that remains is the 648-subframe (one minute) timeout in the
Subframe Lock state.
The other mode fields (similar to Reference Subcarrier ID) between
the Initial Subframe Found and Subframe Lock can be checked for
consistency and a reacquisition can be performed if they are
inconsistent.
While the invention has been described in terms of various
embodiments, it will be apparent to those skilled in the art that
numerous changes can be made to the disclosed embodiments without
departing from the scope of the claims set forth below. For
example, those skilled in the art will understand that the
functions and processes described herein can be implemented using
known circuit components and/or one or more processors programmed
to perform the functions or processes.
* * * * *